Line fault detection method

Abstract

The invention relates to the technical field of electric power, and discloses a line fault detection method, which comprises the following steps of: enabling a first time-varying coefficient and a second time-varying coefficient in a polynomial model to be 0, taking a square of an absolute value as a residual model, and dynamically updating the polynomial model according to the constructed residual model, according to the method, the dynamic nonlinear influence of the mutual inductor on the time domain of the measured signal can be effectively eliminated, and the real fault information can be accurately restored, so that the detection precision of the hidden fault of the overhead line is improved.

Description

Technical Field

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

[0002] With the continuous development of power systems, overhead distribution lines, as a key component of the power supply network, are receiving increasing attention for their operational safety and reliability. However, in actual operation, overhead distribution lines are subject to both external environmental and intrinsic line factors, leading to frequent faults such as high-resistance ground faults and hidden faults. Traditional fault detection methods rely primarily on direct measurement and analysis of line electrical quantities. However, in complex electromagnetic environments, signals often exhibit nonlinear distortion, making accurate fault signature extraction extremely difficult and resulting in low detection accuracy for hidden faults.

[0003] As a commonly used signal acquisition device in overhead lines, the non-ideal characteristics of the mutual inductor have a dynamic nonlinear effect on the time domain of the measured signal. This dynamic nonlinear effect will change with time and environmental changes, causing fault signal distortion, making it impossible for the detection system to accurately restore the true fault information, thereby affecting the accuracy of fault detection. Summary of the Invention

[0004] Based on this, it is necessary to propose a line fault detection method to address the above problems, which can effectively eliminate the dynamic nonlinear influence of the mutual inductor on the time domain of the measured signal, accurately restore the real fault information, and thus improve the detection accuracy of hidden faults of overhead lines.

[0005] To achieve the above object, the present invention provides a line fault detection method in a first aspect, the method comprising: Obtaining a current distortion signal and a historical distortion signal of the mutual inductor, as well as an experimental restoration signal corresponding to the historical distortion signal; Constructing a polynomial model according to the historical distorted signal and the experimental restored signal; Setting the first time-varying coefficient and the second time-varying coefficient in the polynomial model to 0, and taking the square of the absolute value as a residual model, substituting the data point at the tth moment in the current distorted signal into the polynomial model to obtain the data point at the tth moment in the target restored signal, and updating each time-varying coefficient in the polynomial model based on the residual model, the data point at the tth moment in the current distorted signal, and each time-varying coefficient in the polynomial model to obtain an updated polynomial model, and using the updated polynomial model as the polynomial model; wherein t successively takes integers greater than 0 until t equals the total number of moments of the current distorted signal, to obtain the target restored signal; Multi-feature extraction is performed on the target restoration signal to obtain a signal feature vector, and the signal feature vector is input into a preset machine learning model to obtain a fault classification detection result.

[0006] Optionally, updating each time-varying coefficient in the polynomial model according to the residual model, the data point at the t-th moment in the current distorted signal, and each time-varying coefficient in the polynomial model to obtain an updated polynomial model includes: Substituting the data point at the t-th moment in the current distorted signal into the residual model to obtain the residual at the t-th moment; Determining the nth time-varying coefficient at time t+1 based on the residual at time t, the data point at time t in the current distorted signal, and the nth time-varying coefficient in the polynomial model; wherein n successively takes integers greater than -1 until n equals the total number of time-varying coefficients, thereby obtaining the respective time-varying coefficients at time t+1; According to each time-varying coefficient at time t+1, each time-varying coefficient in the polynomial model is updated to obtain the updated polynomial model.

[0007] Optionally, substituting the data point at the t-th moment in the current distorted signal into the residual model to obtain the residual at the t-th moment includes: Using the expression of the residual model Get the residual at the tth moment; in, is the residual at the tth moment, is the n+1th time-varying coefficient, is the total number of time-varying coefficients, is the data point at the tth moment in the current distorted signal.

[0008] Optionally, determining the nth time-varying coefficient at time t+1 based on the residual at time t, the data point at time t in the current distorted signal, and the nth time-varying coefficient in the polynomial model includes: Using the formula Determine the nth time-varying coefficient at time t+1; in, is the nth time-varying coefficient at time t+1, is the n+1th time-varying coefficient, is the preset learning rate, is the residual at the tth moment, is the data point at the tth moment in the current distorted signal.

[0009] Optionally, substituting the data point at the t-th moment in the current distorted signal into the polynomial model to obtain the data point at the t-th moment in the target restored signal includes: The expression using the polynomial model

[0010] determining the target restoration signal; in, The data point at the tth moment in the target restoration signal is: is the n+1th time-varying coefficient, is the total number of time-varying coefficients, is the data point at the tth moment in the current distorted signal.

[0011] Optionally, constructing a polynomial model according to the historical distorted signal and the experimental restoration signal includes: performing standardization processing on the historical distorted signal and the experimental restored signal respectively to obtain a standard distorted signal and a standard restored signal; Determine a plurality of time-varying coefficients based on the data points at each moment in the standard distorted signal and the data points at each moment in the standard restored signal using a least squares method; The polynomial model is constructed according to the various time-varying coefficients.

[0012] Optionally, after constructing the polynomial model according to the historical distorted signal and the experimental restored signal, the method further includes: Performing Fourier transform on the current distorted signal, the historical distorted signal, and the experimental restored signal respectively to obtain a current frequency domain distorted signal, a historical frequency domain distorted signal, and an experimental frequency domain restored signal; Constructing a frequency response transfer function according to the historical frequency domain distortion signal and the experimental frequency domain restoration signal; transforming the frequency response transfer function to obtain a frequency domain form function, and determining a target frequency domain restoration signal based on an amplitude attenuation coefficient and a phase offset of the frequency domain form function and the amplitude and phase of each frequency in the current frequency domain distorted signal; Performing an inverse Fourier transform on the target frequency domain restored signal to obtain an initial restored signal; The initial restored signal is used as the current distorted signal.

[0013] Optionally, the frequency response transfer function is expressed as: ; in, ; In the above formula, is the frequency response transfer function, is the Laplace operator, is the m+1th molecular transfer coefficient, is the total number of numerator transfer coefficients or denominator transfer coefficients, is the m+1th denominator transfer coefficient, is the imaginary unit, is pi, is the frequency.

[0014] Optionally, the frequency domain form function is expressed as: ; in, is the frequency domain form function, is the amplitude attenuation coefficient of frequency f, is a natural constant, is the imaginary unit, is the phase offset of frequency f.

[0015] Optionally, determining a target frequency domain restoration signal according to the amplitude attenuation coefficient and phase offset of the frequency domain form function and the amplitude and phase of each frequency in the current frequency domain distorted signal includes: Using the formula Determining the initial frequency domain restored signal; in, The frequency domain component of the kth frequency in the target frequency domain restoration signal, is the kth frequency, is the amplitude of the kth frequency in the current frequency-domain distorted signal, is the frequency f k The amplitude attenuation coefficient, is a natural constant, is the imaginary unit, is the phase of the kth frequency in the current frequency-domain distorted signal, is the frequency f k The phase offset.

[0016] To achieve the above object, the present invention provides, in a second aspect, a line fault detection device, comprising: An acquisition module, configured to acquire a current distortion signal and a historical distortion signal of the mutual inductor, as well as an experimental restoration signal corresponding to the historical distortion signal; A construction module, configured to construct a polynomial model according to the historical distorted signal and the experimental restoration signal; a substitution and update module, configured to set the first time-varying coefficient and the second time-varying coefficient in the polynomial model to 0, and take the square of the absolute value as a residual model, substitute the data point at the tth moment in the current distorted signal into the polynomial model to obtain the data point at the tth moment in the target restored signal, and update each time-varying coefficient in the polynomial model based on the residual model, the data point at the tth moment in the current distorted signal, and each time-varying coefficient in the polynomial model to obtain an updated polynomial model, and use the updated polynomial model as the polynomial model; wherein t successively takes integers greater than 0 until t equals the total number of moments of the current distorted signal, to obtain the target restored signal; The extraction and prediction module is used to perform multi-feature extraction on the target restoration signal to obtain a signal feature vector, and input the signal feature vector into a preset machine learning model to obtain a fault classification detection result.

[0017] To achieve the above objectives, the present invention provides, in a third aspect, a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the line fault detection method as described in any one of the first aspects.

[0018] To achieve the above-mentioned objectives, the present invention provides, in a fourth aspect, a computer device, 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 line fault detection method as described in any one of the first aspects.

[0019] The embodiment of the present invention has the following beneficial effects: the above method obtains the current distortion signal and the historical distortion signal of the mutual inductor, as well as the experimental restoration signal corresponding to the historical distortion signal, and then constructs a polynomial model based on the historical distortion signal and the experimental restoration signal, and then sets the first time-varying coefficient and the second time-varying coefficient in the polynomial model to 0, and takes the square of the absolute value as the residual model, and then substitutes the data point at the tth moment in the current distortion signal into the polynomial model to obtain the data point at the tth moment in the target restoration signal, and updates each time-varying coefficient in the polynomial model based on the residual model, the data point at the tth moment in the current distortion signal and each time-varying coefficient in the polynomial model to obtain an updated polynomial model, and uses the updated polynomial model as the polynomial model, wherein t successively takes integers greater than 0 until t is equal to the total number of moments of the current distortion signal, to obtain the target restoration signal, and finally performs multi-feature extraction on the target restoration signal to obtain the signal feature vector The signal feature vector is input into the preset machine learning model to obtain the fault classification detection result; that is, by setting the first time-varying coefficient and the second time-varying coefficient in the polynomial model to 0, and taking the square of the absolute value as the residual model, and then substituting the data point at the tth moment in the current distorted signal into the polynomial model, the data point at the tth moment in the target restored signal is obtained, and according to the residual model, the data point at the tth moment in the current distorted signal and each time-varying coefficient in the polynomial model, each time-varying coefficient in the polynomial model is updated to obtain an updated polynomial model, and the updated polynomial model is used as the polynomial model. The polynomial model can be dynamically updated according to the constructed residual model, so that the dynamically updated polynomial model can dynamically restore the time domain of the current distorted signal. This method can effectively eliminate the dynamic nonlinear influence of the mutual inductor on the time domain of the measured signal, accurately restore the real fault information, and thus improve the detection accuracy of hidden faults of overhead lines. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] 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.

[0021] in: Figure 1 A schematic diagram of a line fault detection method according to an embodiment of the present application; Figure 2 This is a schematic diagram of a line fault detection device according to an embodiment of the present application; Figure 31 is a diagram of the internal structure of a computer device in some embodiments. DETAILED DESCRIPTION

[0022] 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.

[0023] With the continuous development of power systems, overhead distribution lines, as a key component of the power supply network, are receiving increasing attention for their operational safety and reliability. However, in actual operation, overhead distribution lines are subject to both external environmental and intrinsic line factors, leading to frequent faults such as high-resistance ground faults and hidden faults. Traditional fault detection methods rely primarily on direct measurement and analysis of line electrical quantities. However, in complex electromagnetic environments, signals often exhibit nonlinear distortion, making accurate fault signature extraction extremely difficult and resulting in low detection accuracy for hidden faults.

[0024] As a commonly used signal acquisition device in overhead lines, the non-ideal characteristics of the mutual inductor have a dynamic nonlinear effect on the time domain of the measured signal. This dynamic nonlinear effect will change with time and environmental changes, causing fault signal distortion, making it impossible for the detection system to accurately restore the true fault information, thereby affecting the accuracy of fault detection.

[0025] In response to the above problems, the present application proposes a line fault detection method that can effectively eliminate the dynamic nonlinear influence of the mutual inductor on the time domain of the measured signal, accurately restore the real fault information, and thus improve the detection accuracy of hidden faults of overhead lines. The specific implementation principle will be described in detail in the following embodiments.

[0026] It should be noted that the present application has three stages for the nonlinear influence of the non-ideal characteristics of the transformer on the measured signal: the first stage is the nonlinear influence of the non-ideal characteristics of the transformer on the amplitude and phase of the measured signal, which is generally caused by the core magnetic permeability, winding inductance and other characteristics of the transformer; the second stage is the nonlinear influence of the non-ideal characteristics of the transformer on the time domain of the measured signal, which is generally caused by core saturation, hysteresis and other characteristics; the third stage is the dynamic nonlinear influence of the non-ideal characteristics of the transformer on the time domain of the measured signal, which is generally caused by the temperature drift, core aging and other characteristics of the transformer.

[0027] In a first aspect, the present application provides a line fault detection method.

[0028] See also Figure 1 , is a schematic diagram of a line fault detection method in an embodiment of the present application, the method comprising: Step 110: Acquire the current distortion signal and the historical distortion signal of the mutual inductor, as well as the experimental restoration signal corresponding to the historical distortion signal.

[0029] The signal here may be an electrical signal, which includes but is not limited to a current signal, a voltage signal, and the like.

[0030] It should be noted that for the current distortion signal of the transformer, the signal output by the transformer can be the current and / or voltage in the line currently measured in real time; for the historical distortion signal, the experimental restoration signal can be input into the transformer, and then the signal output by the transformer; for the experimental restoration signal, the operator can use the signal generated by the signal generator in advance.

[0031] Regarding the method of generating the experimental restoration signal, in some embodiments, the operating frequency range of the mutual inductor can be obtained, the signal generator can be controlled to generate a signal, and within the operating frequency range, the frequency of the generated signal can be adjusted incrementally according to a preset step size to obtain a signal with multiple frequencies. The signal with multiple frequencies is the experimental restoration signal.

[0032] Regarding the value of the operating frequency range, in some embodiments, the present application preferably sets the operating frequency range to 0.1 Hz to 10 kHz.

[0033] Step 120: Construct a polynomial model based on the historical distorted signal and the experimentally restored signal.

[0034] Regarding the construction method of the polynomial model, in some embodiments, multiple time-varying coefficients can be determined based on data points at various moments in the historical distorted signal and data points at various moments in the experimentally restored signal to construct the polynomial model.

[0035] Among them, for the method of determining multiple time-varying coefficients and for the method of constructing a polynomial model, in some embodiments, a multi-order polynomial can be used. Based on the data points at each moment in the historical distorted signal and the data points at each moment in the experimentally restored signal, multiple coefficients of the multi-order polynomial can be determined, and the multiple coefficients are all used as time-varying coefficients to obtain multiple time-varying coefficients, and then the multiple time-varying coefficients are substituted into the multi-order polynomial to obtain the constructed polynomial model; among them, the order of the multi-order polynomial can be obtained and pre-set by the operator based on a lot of experience, experiments or statistics, and of course, it can also be pre-set by the operator according to actual needs.

[0036] It should be noted that the non-ideal characteristics of the mutual inductor not only have a dynamic nonlinear effect on the time domain of the measured signal, but also have a nonlinear effect on the amplitude and phase of the measured signal. This nonlinear effect of amplitude and phase will also cause the fault signal to be distorted, making it impossible for the detection system to accurately restore the true fault information, thereby affecting the accuracy of fault detection. Therefore, after constructing the polynomial model, that is, after step 120, in other embodiments, the present application can also compensate for the frequency domain distortion of the amplitude and phase of the current distorted signal.

[0037] Step 130: Set the first time-varying coefficient and the second time-varying coefficient in the polynomial model to 0, and take the square of the absolute value as the residual model, substitute the data point at the t-th moment in the current distorted signal into the polynomial model, and obtain the data point at the t-th moment in the target restored signal, and update each time-varying coefficient in the polynomial model according to the residual model, the data point at the t-th moment in the current distorted signal, and each time-varying coefficient in the polynomial model to obtain an updated polynomial model, and use the updated polynomial model as the polynomial model; wherein t successively takes integers greater than 0 until t is equal to the total number of moments of the current distorted signal, so as to obtain the target restored signal.

[0038] It should be noted that due to the non-ideal characteristics of the mutual inductor, the time domain of the measured signal has a dynamic nonlinear effect. In this regard, the present application constructs a polynomial model, then sets the first time-varying coefficient and the second time-varying coefficient in the polynomial model to 0, and takes the square of the absolute value as the residual model, and substitutes the data point at the tth moment in the current distorted signal into the polynomial model to obtain the data point at the tth moment in the target restored signal. According to the residual model, the data point at the tth moment in the current distorted signal and each time-varying coefficient in the polynomial model, each time-varying coefficient in the polynomial model is updated to use the updated polynomial model as the polynomial model again, that is, by dynamically updating the polynomial model, the time domain distortion of the data point at each moment in the current distorted signal is dynamically compensated, so as to obtain the target restored signal through dynamic compensation of the time domain distortion.

[0039] It should be further explained that, since each time-varying coefficient in the polynomial model is updated at each moment, and the residual model is obtained based on the polynomial model, accordingly, each time-varying coefficient in the residual model is also updated as each time-varying coefficient in the polynomial model is updated.

[0040] It should be further explained that, in step 130, in some other embodiments, the present application may also compensate for the time domain distortion of the current distorted signal.

[0041] That is, for the method of determining the target restoration signal, in other embodiments, the data point at the t-th moment in the current distorted signal can be substituted into the polynomial model to obtain the data point at the t-th moment in the target restoration signal, where t successively takes integers greater than 0 until n is equal to the total number of moments of the current distorted signal, thereby obtaining the target restoration signal.

[0042] In the present application, a polynomial model is constructed by solving the time-varying coefficients, and then the data points at each moment in the current distorted signal are sequentially substituted into the polynomial model, and the time domain distortion of the data points at each moment in the current distorted signal is compensated. By compensating for the time domain distortion, the target restored signal is obtained, which can effectively eliminate the nonlinear influence of the mutual inductor on the time domain of the measured signal, accurately restore the real fault information, and thus improve the detection accuracy of hidden faults of overhead lines.

[0043] Regarding the method for determining the data point at time t in the target restored signal, in some embodiments, the expression of the polynomial model can be used: Get the target restoration signal; where, The data point at the tth moment in the target restoration signal is: is the n+1th time-varying coefficient, is the total number of time-varying coefficients, is the data point at the tth moment in the current distorted signal.

[0044] In this application, the data points of the target restored signal at each moment are accurately determined through a polynomial model expression, which effectively improves the accuracy and reliability of signal restoration.

[0045] Step 140: Perform multi-feature extraction on the target restoration signal to obtain a signal feature vector, and input the signal feature vector into a preset machine learning model to obtain a fault classification detection result.

[0046] The preset machine learning model here refers to a pre-trained model used to predict the output fault classification detection results based on the input signal feature vector.

[0047] Regarding the signal feature types extracted from the multiple features contained in the signal feature vector, in some embodiments, the signal feature vector includes the spectrum center frequency, bandwidth, short-time energy, time domain waveform change, pulse index, kurtosis, margin factor, waveform factor, frequency standard deviation, mean square frequency, energy entropy, etc.

[0048] Regarding the method of obtaining the preset machine learning model, in some embodiments, a large number of experimental restoration signals can be obtained, and multi-feature extraction can be performed on the experimental restoration signals to obtain experimental signal feature vectors, and the labels corresponding to the experimental signal feature vectors can be determined, and then the large number of experimental signal feature vectors and their corresponding labels can be input into the initial machine learning model for training to obtain the preset machine learning model; in other embodiments, a large number of historical distorted signals can be obtained (that is, they can be obtained based on experimental restoration signals or obtained from line measurements), and signal restoration can be performed through the above steps 110 to 130 to obtain historical restoration signals, and then multi-feature extraction can be performed on the historical restoration signals to obtain historical signal feature vectors, and the labels corresponding to the historical signal feature vectors can be determined, and finally a large number of historical signal feature vectors and their corresponding labels can be input into the initial machine learning model for training to obtain the preset machine learning model.

[0049] In an embodiment of the present application, the first time-varying coefficient and the second time-varying coefficient in the polynomial model are set to 0, and the square of the absolute value is taken as the residual model, and then the data point at the tth moment in the current distorted signal is substituted into the polynomial model to obtain the data point at the tth moment in the target restored signal, and each time-varying coefficient in the polynomial model is updated according to the residual model, the data point at the tth moment in the current distorted signal and each time-varying coefficient in the polynomial model to obtain an updated polynomial model, and the updated polynomial model is used as the polynomial model. The polynomial model can be dynamically updated according to the constructed residual model, so that the dynamically updated polynomial model can dynamically restore the time domain of the current distorted signal. This method can effectively eliminate the dynamic nonlinear influence of the mutual inductor on the time domain of the measured signal, accurately restore the real fault information, and thus improve the detection accuracy of hidden faults of overhead lines.

[0050] In addition, in addition to the already mentioned ability to effectively eliminate the influence of the transformer on the time domain dynamic nonlinearity of the measurement signal, accurately restore the real fault information, and improve the detection accuracy of hidden faults of overhead lines, this line fault detection method also has the following advantages: Comprehensive consideration of multiple nonlinear influences: The nonlinear influence of the non-ideal characteristics of the transformer on the measurement signal in amplitude, phase, time domain static and dynamic aspects is elaborated in detail, and these factors are comprehensively considered in the method design. After constructing the polynomial model, the current distorted signal is also compensated for the frequency domain distortion of amplitude and phase, as well as the time domain distortion. The distorted signal is processed from multiple dimensions to make the restored target signal closer. Real signal, improving the integrity and accuracy of signal restoration; Dynamic update of polynomial model: By constructing a residual model, and dynamically updating the time-varying coefficients of the polynomial model according to the residual model, the current distorted signal data points and the time-varying coefficients in the polynomial model, this dynamic update mechanism can adapt to the characteristics of the non-ideal characteristics of the mutual inductor that change with time and environment, so that the polynomial model always maintains a good fitting effect, thereby more accurately restoring the distorted signals at different times, and enhancing the applicability of the method in complex and changing environments; Accurately determine the target restored signal data points: Use the polynomial model expression to accurately determine the target restored signal data points at each moment, for subsequent Fault detection provides a high-quality signal foundation, effectively improving the accuracy and reliability of signal restoration, and facilitating more accurate analysis of line fault conditions. Multi-feature extraction improves fault identification capabilities: Multi-feature extraction is performed on the target restored signal, and the resulting signal feature vector contains multiple feature types such as spectral center frequency, bandwidth, and short-term energy. These rich features can reflect signal characteristics from different perspectives, providing more comprehensive and detailed information for fault classification and detection, helping machine learning models to more accurately identify different types of faults and improve the accuracy and reliability of fault detection. Flexible acquisition of preset machine learning models: Two methods are provided for obtaining preset machine learning models: training through experimental restoration signals or training using historical restoration signals obtained by restoring historical distorted signals. This flexibility allows for the selection of appropriate model construction methods based on available data resources and actual conditions in practical applications, improving the practicality and operability of the method. At the same time, the use of a large amount of experimental or historical data for model training helps improve the model's generalization ability, enabling it to achieve good fault detection results in different scenarios. Adaptability to complex electromagnetic environments: Traditional fault detection methods face difficulties in extracting fault features and low detection accuracy in complex electromagnetic environments due to nonlinear signal distortion. This method effectively eliminates the various influences brought about by the non-ideal characteristics of the transformer and restores the real fault information. It can maintain a high fault detection accuracy in complex electromagnetic environments, thereby enhancing the operational stability and reliability of the power system in complex environments.

[0051] In a feasible implementation, step 130 in the above embodiment, updating each time-varying coefficient in the polynomial model based on the residual model, the data point at the t-th moment in the current distorted signal, and each time-varying coefficient in the polynomial model to obtain an updated polynomial model, includes: substituting the data point at the t-th moment in the current distorted signal into the residual model to obtain the residual at the t-th moment; determining the n-th time-varying coefficient at the t+1-th moment based on the residual at the t-th moment, the data point at the t-th moment in the current distorted signal, and the n-th time-varying coefficient in the polynomial model; wherein n successively takes integers greater than -1 until n equals the total number of time-varying coefficients, to obtain each time-varying coefficient at the t+1-th moment; and updating each time-varying coefficient in the polynomial model based on each time-varying coefficient at the t+1-th moment to obtain the updated polynomial model.

[0052] In an embodiment of the present application, by dynamically updating the time-varying coefficients of the polynomial model, time-domain dynamic nonlinear compensation of the current distorted signal is achieved, which significantly improves the signal restoration accuracy, thereby completely eliminating the impact of the non-ideal characteristics of the mutual inductor on the detection accuracy.

[0053] It can be understood that the residual drives dynamic adjustment: the current distorted signal is substituted into the residual model (based on the absolute value square of the polynomial model), and the time-varying coefficient of the next moment is dynamically derived by calculating the residual at the current moment, combining the current distorted signal value and the current time-varying coefficient. For example, the residual at the tth moment is used to update the nth time-varying coefficient at the t+1th moment, and iterates successively until all coefficients are updated. The residual model directly reflects the degree of signal distortion, ensures that the adjustment direction of the time-varying coefficient is consistent with the actual error, and avoids the compensation failure of the static model due to environmental changes; the time-varying coefficient is optimized point by point: through a recursive update mechanism (such as n iterating from 0 to the total number of time-varying coefficients), the time-varying coefficient at each moment is optimized based on the previous moment data and the current residual. For example, the nth The update of the time-varying coefficients depends on the residual at the tth moment, the current distorted signal value and the previous coefficients, forming a closed-loop feedback. The coefficient update is matched with the time-varying characteristics of the signal in real time, effectively compensating for the dynamic nonlinear distortion caused by temperature drift, core aging, etc., and improving the adaptability of the model; iterative reconstruction of the polynomial model: the updated time-varying coefficients are substituted into the original polynomial model to generate a new model that adapts to the signal characteristics of the current moment, and is cyclically applied to the data compensation at subsequent moments. For example, after the coefficient update at each moment is completed, the new model is used to process the data at the next moment until the entire signal period is covered. The dynamic reconstruction of the model ensures that the compensation at each moment is based on the latest signal characteristics, eliminates cumulative errors, and the time domain dynamic distortion compensation accuracy of the final restored signal (target restored signal) is significantly better than that of the static model.

[0054] In addition, this update method achieves the ultimate elimination of the influence of the transformer's time-domain dynamic nonlinearity through a three-step closed loop of residual quantization distortion, recursive optimization coefficients, and model dynamic reconstruction, allowing the fault signal restoration accuracy to break through the limitations of traditional static compensation and provide a more reliable data foundation for the detection of hidden faults in overhead lines.

[0055] In a feasible implementation, in the above embodiment, the data point at the t-th moment in the current distorted signal is substituted into the residual model to obtain the residual at the t-th moment, including: Using the expression of the residual model Get the residual at the tth moment; in, is the residual at the tth moment, is the n+1th time-varying coefficient, is the total number of time-varying coefficients, is the data point at the tth moment in the current distorted signal.

[0056] In an embodiment of the present application, the residual at the tth moment is determined by using the residual model expression, which provides a key basis for dynamically updating the time-varying coefficients of the polynomial model, effectively improving the signal restoration accuracy, and thus eliminating the impact of the non-ideal characteristics of the mutual inductor on the detection accuracy.

[0057] It can be understood that in the process of compensating for the influence of the non-ideal characteristics of the mutual inductor on the time domain dynamic nonlinearity of the measurement signal, a residual model is constructed and combined with the polynomial model. The residual at the t-th moment is accurately calculated through the residual model expression. This residual reflects the degree of deviation between the current distorted signal and the actual signal at that moment. Based on this residual value, combined with the data point at the t-th moment in the current distorted signal and each time-varying coefficient in the polynomial model, the various time-varying coefficients at the t+1-th moment can be accurately determined. The polynomial model is dynamically adjusted according to these updated time-varying coefficients, so that the polynomial model can better adapt to the changes of the signal at different moments, thereby performing more accurate time domain distortion dynamic compensation for the data point at each moment in the current distorted signal, and finally obtaining a signal that is closer to the real fault information, which significantly improves the detection accuracy of hidden faults of overhead lines.

[0058] In a feasible implementation, the above embodiment determines the nth time-varying coefficient at time t+1 based on the residual at time t, the data point at time t in the current distorted signal, and the nth time-varying coefficient in the polynomial model, including: Using the formula Determine the nth time-varying coefficient at time t+1; in, is the nth time-varying coefficient at time t+1, is the n+1th time-varying coefficient, is the preset learning rate, is the residual at the tth moment, is the data point at the tth moment in the current distorted signal.

[0059] The preset learning rate may be obtained and pre-set by the operator based on a large amount of experience, experiments or statistics. Of course, it may also be pre-set by the operator based on actual needs.

[0060] In the embodiment of the present application, the adaptive update of the time-varying coefficients of the polynomial model is achieved through a dynamic formula, which significantly improves the accuracy and efficiency of time-domain dynamic nonlinear compensation.

[0061] It can be understood that the dynamic adjustment mechanism: a preset learning rate is introduced into the formula, and the time-varying coefficient is dynamically adjusted by multiplying the residual by the current data point, so that the model can optimize the compensation parameters in real time according to the degree of signal distortion, avoiding slow convergence or oscillation problems; residual-driven optimization: the residual directly reflects the deviation between the model prediction and the actual signal, and as a weight term, ensures that the update direction of the time-varying coefficient always points to the goal of minimizing distortion, thereby improving the targeted compensation; historical data utilization: combining the residual at the previous moment with the current data point to form a feedback loop on the time series, so that the model can capture the trend of the dynamic nonlinear characteristics of the transformer changing with time / environment, and enhance the robustness of the compensation; parameter adaptability: the preset learning rate can be flexibly configured, and the operator can adjust the update sensitivity according to the actual scenario (such as temperature drift speed, core aging degree), balance compensation accuracy and computational efficiency, and meet diverse application needs.

[0062] In addition, this design closely combines the physical layer distortion characteristics with the algorithm layer dynamic compensation through mathematical formulas, breaking through the compensation limitations of traditional static models for dynamic nonlinear effects, and providing a more accurate signal restoration basis for overhead line fault detection.

[0063] In one feasible implementation, step 130 in the above embodiment, substituting the data point at the t-th moment in the current distorted signal into the polynomial model to obtain the data point at the t-th moment in the target restored signal, includes: Expression using polynomial model Determine the target restoration signal; in, The data point at the tth moment in the target restoration signal is: is the n+1th time-varying coefficient, is the total number of time-varying coefficients, is the data point at the tth moment in the current distorted signal.

[0064] In the embodiment of the present application, the data points of the target restored signal at each moment are accurately determined through a polynomial model expression, which significantly improves the accuracy and reliability of time domain dynamic nonlinear compensation and eliminates the influence of the non-ideal characteristics of the mutual inductor on the detection accuracy to the extreme.

[0065] It can be understood that dynamic compensation for time-varying characteristics: the polynomial model dynamically updates the time-varying coefficients to track the changes in the non-ideal characteristics of the mutual inductor over time and the environment in real time, ensuring that the data points at each moment are compensated based on the current optimal model, avoiding the error accumulation caused by the characteristic drift of the static model; residual-driven adaptive optimization: the residual model is used to quantify the compensation error, and the time-varying coefficients are dynamically adjusted in combination with the preset learning rate, so that the model parameters gradually approach the true value, significantly reducing the impact of dynamic nonlinear distortion in the time domain and improving the accuracy of signal restoration; full-process closed-loop correction: from initial signal acquisition to final restoration, a complete signal restoration closed loop is formed through the layered processing of frequency domain distortion compensation, time domain static compensation and time domain dynamic compensation, ensuring that the final output target restoration signal is highly close to the real fault information, providing a reliable basis for fault detection.

[0066] In a feasible implementation, step 120 in the above embodiment, constructing a polynomial model based on the historical distorted signal and the experimental restoration signal, includes: standardizing the historical distorted signal and the experimental restoration signal respectively to obtain a standard distorted signal and a standard restoration signal; using the least squares method to determine multiple time-varying coefficients based on the data points at each moment in the standard distorted signal and the data points at each moment in the standard restoration signal; and constructing a polynomial model based on each time-varying coefficient.

[0067] Regarding the method for determining the standard distorted signal and the standard restored signal, in some embodiments, the historical distorted signal and the experimental restored signal can be standardized respectively to obtain the standard distorted signal and the standard restored signal; a first mean and a first standard deviation of the historical distorted signal are determined, and a second mean and a second standard deviation of the experimental restored signal are determined; then, according to the first mean and the first standard deviation, the data points at each moment in the historical distorted signal are standardized to obtain the standard distorted signal; and according to the second mean and the second standard deviation, the data points at each moment in the experimental restored signal are standardized to obtain the standard restored signal.

[0068] In this application, the standardization processing based on mean and standard deviation is used to improve the standardization and comparability of signal data, laying a solid foundation for the subsequent accurate determination of time-varying coefficients and construction of polynomial models.

[0069] In an embodiment of the present application, the time-varying coefficients of the polynomial model are determined by standardization processing and the use of the least squares method, thereby improving the efficiency and accuracy of model construction, thereby improving the effect of time domain distortion compensation or time domain distortion dynamic compensation.

[0070] It can be understood that the historical distorted signal and the experimental restoration signal are standardized to obtain the standard distorted signal and the standard restoration signal respectively. The standardization process can eliminate the dimensional influence and outlier interference in the signal, making the signal data more standardized and unified, and providing a more stable data basis for subsequent model construction; the least squares method is used to determine multiple time-varying coefficients based on the data points at each moment in the standard distorted signal and the standard restoration signal. The least squares method is a classic mathematical optimization method that finds the best function match for the data by minimizing the sum of squares of errors. It can effectively process the noise and errors in the signal data, making the determined time-varying coefficients more accurate and reliable; a polynomial model is constructed based on these accurate time-varying coefficients. Since the accuracy of the time-varying coefficients is guaranteed, the constructed polynomial model can more accurately describe the relationship between the historical distorted signal and the experimental restoration signal of the transformer. Therefore, when the target restoration signal is subsequently subjected to time domain distortion compensation or time domain distortion dynamic compensation, the nonlinear influence or dynamic nonlinearity of the non-ideal characteristics of the transformer on the signal time domain can be more effectively eliminated, thereby improving the accuracy of fault detection.

[0071] In a feasible implementation, in step 120 of the above embodiment, after constructing the polynomial model based on the historical distorted signal and the experimental restored signal, the method further includes: performing Fourier transform on the current distorted signal, the historical distorted signal, and the experimental restored signal, respectively, to obtain the current frequency domain distorted signal, the historical frequency domain distorted signal, and the experimental frequency domain restored signal; constructing a frequency response transfer function based on the historical frequency domain distorted signal and the experimental frequency domain restored signal; transforming the frequency response transfer function to obtain a frequency domain form function, and determining the target frequency domain restored signal based on the amplitude attenuation coefficient and phase offset of the frequency domain form function, and the amplitude and phase of each frequency in the current frequency domain distorted signal; performing an inverse Fourier transform on the target frequency domain restored signal to obtain an initial restored signal; and using the initial restored signal as the current distorted signal.

[0072] It should be noted that the frequency response transfer function of the present application belongs to the s domain of the Laplace transform, and the frequency form function belongs to the frequency domain of the Fourier transform. Therefore, in some embodiments, the frequency response transfer function can be converted between the s domain and the frequency domain to obtain the frequency domain form function; where s is the Laplace operator.

[0073] It should be further explained that due to the non-ideal characteristics of the mutual inductor, the amplitude and phase of the measured signal have a nonlinear effect. In this regard, the present application compensates for the frequency domain distortion of the amplitude and phase of each frequency in the current frequency domain distorted signal according to the amplitude attenuation coefficient and phase offset of the frequency domain form function, so as to obtain the target frequency domain restoration signal through the compensation of the frequency domain distortion.

[0074] It should be further explained that the current distorted signal initially measured by the mutual inductor of the present application belongs to the time domain signal, and the target frequency domain restoration signal belongs to the frequency domain signal. Therefore, in some embodiments, the target frequency domain restoration signal can be subjected to an inverse Fourier transform in the frequency domain and time domain to obtain the initial restoration signal in the time domain.

[0075] Regarding the method of constructing the frequency response transfer function, in some embodiments, the frequency response of each frequency can be determined based on the frequency domain component of each frequency in the historical frequency domain distortion signal and the frequency domain component of each frequency in the experimental frequency domain restoration signal, and based on the frequency response of each frequency and the preset objective function, multiple numerator transfer coefficients and multiple denominator transfer coefficients are obtained to construct the frequency response transfer function; wherein, the preset objective function can be obtained and pre-set by the operator based on a large amount of experience, experiments or statistics, and of course, it can also be pre-set by the operator based on actual needs.

[0076] It should be noted that the purpose of the preset objective function is to obtain the corresponding multiple numerator transfer coefficients and multiple denominator transfer coefficients when the frequency response of each frequency is minimized, so that a frequency response transfer function can be constructed based on the multiple numerator transfer coefficients and multiple denominator transfer coefficients.

[0077] Furthermore, with respect to the method of constructing the preset objective function and the method of constructing the frequency response transfer function, in some embodiments, a transfer function can be constructed according to multi-order polynomials in both the numerator and the denominator, and then a preset objective function can be constructed based on the transfer function and the frequency response of each frequency; wherein the frequency response of each frequency is known, and in the transfer function, only the multiple coefficients in the multi-order polynomial of the numerator and the multiple coefficients in the multi-order polynomial of the denominator are unknown and adjustable. Therefore, by adjusting the multiple coefficients in the multi-order polynomial of the numerator and the multiple coefficients in the multi-order polynomial of the denominator so that the frequency response of each frequency reaches the minimum, the multiple coefficients in the multi-order polynomial of the numerator adjusted this time are all used as numerator transfer coefficients, and the multiple coefficients in the multi-order polynomial of the denominator are all used as denominator transfer coefficients, and the multiple numerator transfer coefficients and the multiple denominator transfer coefficients are substituted into the transfer function to obtain the constructed frequency response transfer function.

[0078] Furthermore, with respect to the method for constructing a frequency response transfer function, in some embodiments, the quotient between the frequency domain component of the dth frequency in the historical frequency domain distortion signal and the frequency domain component of the dth frequency in the experimental frequency domain restoration signal can be taken as the frequency response of the dth frequency, where d successively takes integers greater than 0 until d is equal to the total number of frequencies of the historical frequency domain distortion signal or the experimental frequency domain restoration signal, and the frequency response of each frequency is obtained. Then, using the LM method, while satisfying a preset objective function, the numerator transfer coefficients and the denominator transfer coefficients in the preset objective function are adjusted according to the frequency response of each frequency, so that the function value of the preset objective function is minimized, and multiple numerator transfer coefficients and multiple denominator transfer coefficients are obtained. Finally, a frequency response transfer function is constructed based on the numerator transfer coefficients and the denominator transfer coefficients; where LM is Levenberg-Marquardt, also known as the Levenberg-Marquardt method (also known as the Levenberg-Marquardt method).

[0079] In the embodiment of the present application, the construction process of the frequency response transfer function is optimized by the LM method, which significantly improves the convergence speed and accuracy of the transfer coefficient solution, and at the same time enhances the model's adaptability to complex nonlinear distortion, providing a more reliable foundation for subsequent signal restoration and fault detection.

[0080] Furthermore, the expression of the preset objective function is: ;in, ; In the above formula, is the function value of the preset objective function, is the total number of frequencies of the historical frequency domain distorted signal or the experimental frequency domain restored signal, is the frequency response of the dth frequency, is the Laplace operator corresponding to the dth frequency, is the m+1th molecular transfer coefficient, is the total number of numerator transfer coefficients or denominator transfer coefficients, is the m+1th denominator transfer coefficient, is the imaginary unit, is pi, is the dth frequency.

[0081] In an embodiment of the present application, by constructing a composite objective function based on frequency response, the fitting accuracy of the transfer function to the non-ideal characteristics of the mutual inductor is significantly improved, while the anti-interference ability of the model in complex electromagnetic environments is enhanced, providing a more reliable mathematical basis for subsequent signal restoration.

[0082] In an embodiment of the present application, the frequency response of each frequency is determined based on the frequency domain components of each frequency in the historical frequency domain distortion signal and the frequency domain components of each frequency in the experimental frequency domain restoration signal, and multiple numerator transfer coefficients and multiple denominator transfer coefficients are obtained based on the frequency responses of each frequency and the preset objective function to construct a frequency response transfer function, so that the amplitude and phase of the current frequency domain distortion signal are restored according to the amplitude attenuation coefficient and phase offset of the frequency domain transfer function corresponding to the constructed frequency response transfer function. This can not only effectively eliminate the nonlinear influence of the mutual inductor on the time domain of the measured signal, as well as the dynamic nonlinear influence in the time domain, but also further eliminate its nonlinear influence on the amplitude and phase of the measured signal, thereby extremely accurately restoring the real fault information, thereby improving the detection accuracy of hidden faults of overhead lines.

[0083] In a feasible implementation, the frequency response transfer function in the above embodiment is expressed as: ; in, ; In the above formula, is the frequency response transfer function, is the Laplace operator, is the m+1th molecular transfer coefficient, is the total number of numerator transfer coefficients or denominator transfer coefficients, is the m+1th denominator transfer coefficient, is the imaginary unit, is pi, is the frequency.

[0084] In an embodiment of the present application, by constructing a frequency response transfer function based on the Laplace domain, high-precision mathematical modeling of the non-ideal characteristics of the mutual inductor is achieved, which significantly improves the amplitude / phase accuracy of signal restoration, and at the same time enhances the model's adaptability to nonlinear distortion, providing a reliable signal basis for subsequent fault feature extraction.

[0085] It can be understood that the full-band dynamic compensation capability: the transfer function adopts a multi-order polynomial structure, which can accurately fit the amplitude-frequency characteristics and phase-frequency characteristics of the mutual inductor within the operating frequency range; the complex domain parameter optimization mechanism: by introducing imaginary units and Laplace operators, the transfer function can simultaneously optimize the real part (amplitude response) and imaginary part (phase response).

[0086] In a feasible implementation, the frequency domain form function in the above embodiment is expressed as: ; in, is the frequency domain form function, is the amplitude attenuation coefficient of frequency f, is a natural constant, is the imaginary unit, is the phase shift of frequency f.

[0087] In the embodiment of the present application, an independent compensation mechanism for amplitude attenuation and phase shift is realized by constructing a frequency domain formal function, which significantly improves the accuracy and flexibility of frequency domain distortion compensation. At the same time, it provides high-precision initial conditions for the subsequent process, effectively solving the distortion problem caused by the non-ideal characteristics of the mutual inductor.

[0088] It can be understood that the amplitude-phase decoupling compensation mechanism: the frequency domain formal function decouples the amplitude attenuation coefficient and the phase offset into independent parameters, which can accurately compensate for the amplitude-frequency characteristic distortion and phase-frequency characteristic distortion of the mutual inductor respectively; complex domain dynamic modeling capability: by introducing natural constants and imaginary units, the frequency domain formal function can be expressed in complex form, thereby simultaneously characterizing the real part (amplitude) and imaginary part (phase) of the signal. This modeling method enables the model to accurately match the nonlinear offset caused by the mutual inductor; frequency band adaptive compensation characteristics: the amplitude attenuation coefficient and phase offset are both functions of frequency, which can dynamically adapt to the differences in the nonlinear characteristics of the mutual inductor in different frequency bands.

[0089] In one feasible implementation, the above embodiment determines the target frequency domain restoration signal based on the amplitude attenuation coefficient and phase offset of the frequency domain form function and the amplitude and phase of each frequency in the current frequency domain distorted signal, including: Using the formula Determine an initial frequency domain restoration signal; in, The frequency domain component of the kth frequency in the target frequency domain restoration signal is is the kth frequency, is the amplitude of the kth frequency in the current frequency domain distortion signal, is the frequency f k The amplitude attenuation coefficient, is a natural constant, is the imaginary unit, is the phase of the kth frequency in the current frequency domain distortion signal, is the frequency f k The phase offset.

[0090] In the embodiment of the present application, accurate and independent compensation for the amplitude attenuation and phase shift caused by the mutual inductor is achieved through the frequency domain component compensation formula, which significantly improves the restoration accuracy of the frequency domain distorted signal and provides a high-fidelity signal basis for subsequent fault feature extraction.

[0091] It can be understood that the amplitude-phase decoupling compensation mechanism: the formula decouples the amplitude compensation term from the phase compensation term, and can independently optimize the amplitude-frequency characteristic distortion and phase-frequency characteristic distortion of the mutual inductor: complex domain dynamic modeling capability: by introducing natural constants and imaginary units, the formula can be expressed in complex form, thereby simultaneously characterizing the real part (amplitude) and imaginary part (phase) of the signal. This modeling method enables the model to accurately match the nonlinear offset caused by the mutual inductor; frequency band adaptive compensation characteristics: the amplitude attenuation coefficient and phase offset are both functions of frequency, which can dynamically adapt to the differences in the nonlinear characteristics of the mutual inductor in different frequency bands.

[0092] In a feasible implementation method, the above embodiment performs Fourier transform on the current distorted signal to obtain the current frequency domain distorted signal, including: performing Fourier transform on the current distorted signal to obtain an initial frequency domain distorted signal; determining the maximum frequency based on the experimental restoration signal; determining the cutoff frequency based on the maximum frequency and a preset safety factor; and performing frequency screening on the initial frequency domain distorted signal based on the cutoff frequency to obtain the current frequency domain distorted signal.

[0093] The preset safety factor may be obtained and pre-set by the operator based on a large amount of experience, experiments or statistics. Of course, it may also be pre-set by the operator based on actual needs.

[0094] Regarding the value of the preset safety factor, in some embodiments, the present application preferably sets the preset safety factor to 1.2.

[0095] Regarding the method for determining the maximum frequency, in some embodiments, the experimental restoration signal can be Fourier transformed to obtain an experimental frequency domain restoration signal, and then the largest frequency in the frequency domain components of each frequency of the experimental frequency domain restoration signal can be used as the maximum frequency; in other implementations, the maximum frequency can also be determined based on the operating frequency range of the mutual inductor, such as when the operating frequency range of the mutual inductor is 0.1Hz to 10kHz, 10kHz is used as the maximum frequency.

[0096] Regarding the method for determining the cutoff frequency, in some embodiments, the product of the maximum frequency and a preset safety factor may be used as the cutoff frequency.

[0097] Regarding the method for determining the current frequency domain distortion signal, in some embodiments, the frequency domain components of each frequency of the initial frequency domain distortion signal with a frequency greater than a cutoff frequency may be removed to obtain the current frequency domain distortion signal.

[0098] In the embodiment of the present application, high-frequency noise interference is effectively suppressed through a frequency screening mechanism, which significantly improves the signal-to-noise ratio of the fault signal and the accuracy of subsequent feature extraction.

[0099] It can be understood that the adaptive frequency band constraint: the maximum frequency is determined according to the experimental frequency domain restoration signal, and the cutoff frequency is set in combination with a 1.2 times safety factor, which not only retains the effective fault characteristic frequency band, but also avoids the pollution of high-frequency noise on the signal. It is more adaptable than the fixed frequency band solution; the anti-aliasing effect is improved: by accurately eliminating the components outside the cutoff frequency after Fourier transform, it effectively prevents high-frequency noise from aliasing into the baseband frequency band, and provides a purer input signal for subsequent frequency domain distortion compensation; transformer characteristic adaptation: according to the differences in the operating frequency bands of different transformers, the cutoff frequency mechanism can automatically match the equipment characteristics, avoiding the problem of feature loss or noise introduction caused by fixed frequency band.

[0100] In a feasible implementation, step 140 in the above embodiment performs multi-feature extraction on the target restoration signal to obtain a signal feature vector, including: performing a wavelet transform on the target restoration signal to obtain a first time-frequency graph; performing a short-time Fourier transform on the target restoration signal to obtain a second time-frequency graph; performing multi-feature extraction on the first time-frequency graph and the second time-frequency graph to obtain a spectrum center frequency, a frequency bandwidth, a short-time energy, and a time-domain waveform variation; and using the spectrum center frequency, frequency bandwidth, short-time energy, and the time-domain waveform variation as the signal feature vector.

[0101] In the embodiment of the present application, the comprehensiveness of fault signal feature representation and the accuracy of fault classification detection are significantly improved by fusing and extracting features using a multi-time-frequency analysis method.

[0102] It can be understood that the time-frequency joint characterization is enhanced: combining the complementary characteristics of wavelet transform and short-time Fourier transform, it not only retains the time-frequency localization analysis capability of wavelet transform, but also utilizes the frequency resolution advantage of short-time Fourier transform to construct a more complete signal time-frequency feature space; multi-dimensional feature fusion: by extracting the four core features of spectrum center frequency, bandwidth, short-time energy and time domain waveform variation, a multi-dimensional feature vector covering frequency domain, time domain and energy domain is formed, which effectively solves the problem of insufficient representation of complex fault modes by a single feature; anti-interference ability is improved: the time-frequency analysis method has Natural adaptability: noise interference can be suppressed through time-frequency graph processing, and the extracted features such as bandwidth and time-domain waveform change are more sensitive to hidden faults; feature interpretability is enhanced: the four extracted features have clear physical meanings and are strongly correlated with the fault type, providing interpretable input features for the machine learning model, which helps to improve the credibility of the fault classification results; computational efficiency is optimized: a standardized time-frequency transformation process is adopted, and both wavelet transform and short-time Fourier transform can be implemented through fast algorithms. The feature extraction process maintains linear complexity, meeting real-time detection needs while ensuring feature quality.

[0103] In a second aspect, the present application provides a line fault detection device.

[0104] See also Figure 2 , is a schematic diagram of a line fault detection device according to an embodiment of the present application, wherein the device 210 includes: An acquisition module 211 is configured to acquire a current distortion signal and a historical distortion signal of the mutual inductor, as well as an experimental restoration signal corresponding to the historical distortion signal; A construction module 212 is used to construct a polynomial model based on the historical distorted signal and the experimental restoration signal; a substitution and update module 213 configured to set the first time-varying coefficient and the second time-varying coefficient in the polynomial model to 0, and to take the square of the absolute value as a residual model, substitute the data point at the tth moment in the current distorted signal into the polynomial model to obtain the data point at the tth moment in the target restored signal, and update each time-varying coefficient in the polynomial model based on the residual model, the data point at the tth moment in the current distorted signal, and each time-varying coefficient in the polynomial model to obtain an updated polynomial model, and use the updated polynomial model as the polynomial model; wherein t successively takes integers greater than 0 until t equals the total number of moments in the current distorted signal, to obtain the target restored signal; The extraction and prediction module 214 is used to perform multi-feature extraction on the target restoration signal to obtain a signal feature vector, and input the signal feature vector into a preset machine learning model to obtain a fault classification detection result.

[0105] In the embodiment of the present application, the relevant contents of the acquisition module 211, the construction module 212, the substitution and update module 213 and the extraction and prediction module 214 can be found in Figure 1 The contents of the illustrated embodiments are not described in detail here.

[0106] 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.

[0107] In an embodiment of the present application, the first time-varying coefficient and the second time-varying coefficient in the polynomial model are set to 0, and the square of the absolute value is taken as the residual model, and then the data point at the tth moment in the current distorted signal is substituted into the polynomial model to obtain the data point at the tth moment in the target restored signal, and each time-varying coefficient in the polynomial model is updated according to the residual model, the data point at the tth moment in the current distorted signal and each time-varying coefficient in the polynomial model to obtain an updated polynomial model, and the updated polynomial model is used as the polynomial model. The polynomial model can be dynamically updated according to the constructed residual model, so that the dynamically updated polynomial model can dynamically restore the time domain of the current distorted signal. The device can effectively eliminate the dynamic nonlinear influence of the mutual inductor on the time domain of the measured signal, accurately restore the real fault information, and thus improve the detection accuracy of hidden faults of overhead lines.

[0108] In addition, in addition to the already mentioned ability to effectively eliminate the influence of the transformer on the time domain dynamic nonlinearity of the measurement signal, accurately restore the real fault information, and improve the detection accuracy of hidden faults of overhead lines, the line fault detection device also has the following advantages: Comprehensive consideration of multiple nonlinear influences: Detailed explanation of the nonlinear influence of the non-ideal characteristics of the transformer on the measurement signal in amplitude, phase, time domain static and dynamic aspects, and comprehensive consideration of these factors in the design of the device. After constructing the polynomial model, the current distorted signal is also compensated for the frequency domain distortion of amplitude and phase, as well as the time domain distortion. The distorted signal is processed from multiple dimensions to make the restored target signal closer. Real signal, improving the integrity and accuracy of signal restoration; Dynamic update of polynomial model: By constructing a residual model, and dynamically updating the time-varying coefficients of the polynomial model according to the residual model, the current distorted signal data point and the time-varying coefficients in the polynomial model, this dynamic update mechanism can adapt to the characteristics of the non-ideal characteristics of the mutual inductor that change with time and environment, so that the polynomial model always maintains a good fitting effect, thereby more accurately restoring the distorted signal at different times, and enhancing the applicability of the device in complex and changing environments; Accurately determine the target restoration signal data point: Use the polynomial model expression to accurately determine the target restoration signal data point at each moment, for subsequent Fault detection provides a high-quality signal foundation, effectively improving the accuracy and reliability of signal restoration, and facilitating more accurate analysis of line fault conditions. Multi-feature extraction improves fault identification capabilities: Multi-feature extraction is performed on the target restored signal, and the resulting signal feature vector contains multiple feature types such as spectral center frequency, bandwidth, and short-term energy. These rich features can reflect the characteristics of the signal from different angles, providing more comprehensive and detailed information for fault classification and detection, helping machine learning models to more accurately identify different types of faults and improve the accuracy and reliability of fault detection. Flexible acquisition of preset machine learning models: Two methods are provided for obtaining preset machine learning models: training through experimental restoration signals or training using historical restoration signals obtained by restoring historical distorted signals. This flexibility allows for the selection of appropriate model construction methods based on available data resources and actual conditions in practical applications, improving the practicality and operability of the device. At the same time, the use of a large amount of experimental or historical data for model training helps improve the model's generalization ability, enabling it to achieve good fault detection results in different scenarios. Adaptation to complex electromagnetic environments: Traditional fault detection devices face difficulties in extracting fault features and low detection accuracy in complex electromagnetic environments due to nonlinear signal distortion. This device effectively eliminates the various influences caused by the non-ideal characteristics of the transformer and restores the real fault information. It can maintain a high fault detection accuracy in complex electromagnetic environments, thereby enhancing the operating stability and reliability of the power system in complex environments.

[0109] In a third aspect, the present application provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the line fault detection method as described in any one of the first aspects.

[0110] In a fourth aspect, the present application provides a computer device, 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 line fault detection method as described in any one of the first aspects.

[0111] 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.

[0112] 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.

[0113] 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.

[0114] 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).

[0115] 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.

[0116] 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 line fault detection method, characterized in that: The method comprises: Obtaining a current distortion signal and a historical distortion signal of the mutual inductor, as well as an experimental restoration signal corresponding to the historical distortion signal; Constructing a polynomial model according to the historical distorted signal and the experimental restored signal; Setting the first time-varying coefficient and the second time-varying coefficient in the polynomial model to 0, and taking the square of the absolute value as a residual model, substituting the data point at the tth moment in the current distorted signal into the polynomial model to obtain the data point at the tth moment in the target restored signal, and updating each time-varying coefficient in the polynomial model based on the residual model, the data point at the tth moment in the current distorted signal, and each time-varying coefficient in the polynomial model to obtain an updated polynomial model, and using the updated polynomial model as the polynomial model; wherein t successively takes integers greater than 0 until t equals the total number of moments of the current distorted signal, to obtain the target restored signal; Multi-feature extraction is performed on the target restoration signal to obtain a signal feature vector, and the signal feature vector is input into a preset machine learning model to obtain a fault classification detection result.

2. The line fault detection method according to claim 1, characterized in that: The updating of each time-varying coefficient in the polynomial model according to the residual model, the data point at the t-th moment in the current distorted signal, and each time-varying coefficient in the polynomial model to obtain an updated polynomial model includes: Substituting the data point at the t-th moment in the current distorted signal into the residual model to obtain the residual at the t-th moment; Determining the nth time-varying coefficient at time t+1 based on the residual at time t, the data point at time t in the current distorted signal, and the nth time-varying coefficient in the polynomial model; wherein n successively takes integers greater than -1 until n equals the total number of time-varying coefficients, thereby obtaining the respective time-varying coefficients at time t+1; According to each time-varying coefficient at time t+1, each time-varying coefficient in the polynomial model is updated to obtain the updated polynomial model.

3. The line fault detection method according to claim 2, characterized in that: Substituting the data point at the t-th moment in the current distorted signal into the residual model to obtain the residual at the t-th moment includes: Using the expression of the residual model Get the residual at the tth moment; in, is the residual at the tth moment, is the n+1th time-varying coefficient, is the total number of time-varying coefficients, is the data point at the tth moment in the current distorted signal.

4. The line fault detection method according to claim 2, characterized in that: The determining, based on the residual at the t-th moment, the data point at the t-th moment in the current distorted signal, and the n-th time-varying coefficient in the polynomial model, comprises: Using the formula Determine the nth time-varying coefficient at time t+1; in, is the nth time-varying coefficient at time t+1, is the n+1th time-varying coefficient, is the preset learning rate, is the residual at the tth moment, is the data point at the tth moment in the current distorted signal.

5. The line fault detection method according to claim 1, characterized in that: Substituting the data point at the t-th moment in the current distorted signal into the polynomial model to obtain the data point at the t-th moment in the target restored signal includes: The expression using the polynomial model determining the target restoration signal; in, The data point at the tth moment in the target restoration signal is: is the n+1th time-varying coefficient, is the total number of time-varying coefficients, is the data point at the tth moment in the current distorted signal.

6. The line fault detection method according to claim 1, characterized in that: Constructing a polynomial model according to the historical distorted signal and the experimental restoration signal, including: performing standardization processing on the historical distorted signal and the experimental restored signal respectively to obtain a standard distorted signal and a standard restored signal; Determine a plurality of time-varying coefficients based on the data points at each moment in the standard distorted signal and the data points at each moment in the standard restored signal using a least squares method; The polynomial model is constructed according to the various time-varying coefficients.

7. The line fault detection method according to claim 1, characterized in that: After constructing a polynomial model based on the historical distorted signal and the experimental restored signal, the method further includes: Performing Fourier transform on the current distorted signal, the historical distorted signal, and the experimental restored signal respectively to obtain a current frequency domain distorted signal, a historical frequency domain distorted signal, and an experimental frequency domain restored signal; Constructing a frequency response transfer function according to the historical frequency domain distortion signal and the experimental frequency domain restoration signal; transforming the frequency response transfer function to obtain a frequency domain form function, and determining a target frequency domain restoration signal based on an amplitude attenuation coefficient and a phase offset of the frequency domain form function and the amplitude and phase of each frequency in the current frequency domain distorted signal; Performing an inverse Fourier transform on the target frequency domain restored signal to obtain an initial restored signal; The initial restored signal is used as the current distorted signal.

8. The line fault detection method according to claim 7, characterized in that: The expression of the frequency response transfer function is: ； in, ; In the above formula, is the frequency response transfer function, is the Laplace operator, is the m+1th molecular transfer coefficient, is the total number of numerator transfer coefficients or denominator transfer coefficients, is the m+1th denominator transfer coefficient, is the imaginary unit, is pi, is the frequency.

9. The line fault detection method according to claim 7, characterized in that: The expression of the frequency domain form function is: ； in, is the frequency domain form function, is the amplitude attenuation coefficient of frequency f, is a natural constant, is the imaginary unit, is the phase offset of frequency f.

10. The line fault detection method according to claim 7, characterized in that: Determining a target frequency domain restoration signal according to the amplitude attenuation coefficient and the phase offset of the frequency domain form function and the amplitude and phase of each frequency in the current frequency domain distorted signal includes: Using the formula Determining the initial frequency domain restored signal; in, The frequency domain component of the kth frequency in the target frequency domain restoration signal, is the kth frequency, is the amplitude of the kth frequency in the current frequency-domain distorted signal, is the frequency f k The amplitude attenuation coefficient, is a natural constant, is the imaginary unit, is the phase of the kth frequency in the current frequency-domain distorted signal, is the frequency f k The phase offset.

Images