Power distribution line virtual open-circuit fault detection method based on load characteristic separation

Abstract

The application relates to a power distribution line virtual connection and disconnection fault detection method based on load characteristic separation, relates to the technical field of virtual connection and disconnection fault detection, and comprises the power distribution line virtual connection and disconnection fault detection method based on load characteristic separation. The power distribution line is first subjected to high-frequency synchronous sampling to obtain load current and voltage data, then voltage disturbance types and intensities are identified, next, a contaminant effect mathematical model is inquired to obtain a predicted value of a print offset, the offset is used for contaminant compensation on actually measured load characteristics to obtain pure characteristics, and finally, virtual connection and disconnection fault identification is performed based on the pure characteristics and voltage data. The application solves the technical problems that a traditional method is difficult to distinguish between power grid quality disturbance and virtual connection and disconnection fault, disturbance contamination effect leads to invalidation of identification criteria, and the false alarm and missed alarm rates are high.

Description

Technical Field

[0001] This application relates to the field of fault detection technology for loose connections, and in particular to a method for detecting loose connections in distribution lines based on load characteristic separation. Background Technology

[0002] Currently, fault detection methods for loose connections in distribution lines based on load characteristic separation typically utilize non-intrusive load identification technology to extract harmonic and other characteristics of the line current, identifying faults by comparing the measured characteristics with a health status benchmark. However, voltage sags, harmonic distortions, flicker, and other grid quality disturbances commonly found in actual distribution networks can significantly alter the current characteristics of the load.

[0003] Existing technologies, such as Chinese patent CN114676783A, focus on detecting unknown load types, and Chinese patent CN117009737A emphasizes distinguishing early faults from other system disturbances. Neither systematically considers the contamination effect of power grid quality disturbances on the load characteristics themselves. This contamination effect causes the measured load imprint to deviate from the true value, rendering the residual analysis benchmark of the fault identification system ineffective and leading to false alarms or missed alarms.

[0004] Therefore, there is currently a lack of quantitative modeling methods for the pollution effects of power grid quality disturbances, as well as technical means to separate disturbance components from mixed signals and compensate for load characteristics. Summary of the Invention

[0005] This application provides a method for detecting loose connection and open circuit faults in distribution lines based on load characteristic separation. It improves the technical problems in the prior art, such as the difficulty in effectively distinguishing between power grid quality disturbances and loose connection and open circuit faults, as well as the high deployment cost of contact sensors and the high false alarm and false alarm rates of non-intrusive monitoring methods under disturbance and pollution.

[0006] This application discloses the following technical solution: In a first aspect, this application provides a method for detecting loose connection / open circuit faults in power distribution lines based on load characteristic separation, the method comprising: Load current and voltage sampling data obtained through high-frequency sampling on the target power distribution line are loaded; Based on the voltage sampling data, power grid quality disturbance identification is performed to obtain current disturbance characteristic data, wherein the current disturbance characteristic data includes disturbance type and disturbance intensity parameters; Based on the current disturbance characteristic data, a pre-constructed mathematical model of pollutant effects is queried to obtain the predicted imprint offset value; Based on the predicted imprint offset and the sampled load current data, load characteristic pollution compensation is performed to obtain the compensated clean load characteristic data. Based on the pure load characteristic data and the voltage sampling data, a loose connection fault identification is performed to obtain the fault identification result.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: The technical solution of this application provides a method for detecting loose connections and open circuits in power distribution lines based on load feature separation. First, by deploying a sampling device on the target power distribution line, load current and voltage waveform data are collected at high frequency in a time-synchronous manner. This achieves accurate correspondence between voltage and current signals and complete transient waveform capture, providing a time-aligned and sequentially continuous original data foundation for subsequent disturbance identification and load feature extraction. This solves the problems of time misalignment of current and voltage data and loss of fault features due to low sampling rate.

[0008] Furthermore, by performing Fast Fourier Transform and envelope analysis on the voltage sampling data, the total harmonic distortion rate, voltage sag depth, and voltage flicker intensity are calculated, enabling the identification and intensity quantification of three types of power grid quality disturbances: harmonic distortion, voltage sag, and voltage flicker. This provides accurate disturbance parameter inputs for subsequent load characteristic pollution compensation and solves the problem that the types and intensities of power grid quality disturbances cannot be automatically identified and quantified.

[0009] Furthermore, by pre-collecting characteristic data of typical loads under ideal conditions and various disturbance conditions, the measured value of imprint offset is calculated. Using disturbance parameters as input and offset as output, a regression model is trained to construct a mathematical model of pollutant effects. This achieves a quantitative mapping from disturbance characteristics to load imprint offset, providing model support for obtaining predicted imprint offset values ​​in real time and solving the problem of lacking a means to predict load characteristic offsets from disturbance parameters.

[0010] Furthermore, by extracting measured active power, reactive power, total harmonic distortion rate of current, and current waveform coefficient from load current sampling data to form a measured load imprint, and obtaining the corresponding component offset from the imprint offset prediction value, pollution compensation is completed by subtracting the corresponding components. This realizes the restoration of the measured load characteristics polluted by disturbance to the characteristics of a clean load, and solves the problem that grid disturbances cause load characteristics to deviate from the true value, resulting in the failure of the fault identification benchmark.

[0011] Finally, by comparing the pure load characteristics with the benchmark load characteristics to calculate the deviation vector, when the deviation exceeds the threshold, the fault type is matched according to the sign and magnitude of the deviation components. Combined with the topology connection matrix, the fault segment is located by topology backtracking. The benchmark is updated by sliding average during periods of no disturbance and no fault. At the same time, the confidence level is dynamically adjusted by calculating the comprehensive pollution coefficient. This realizes a complete fault identification chain from deviation quantification, fault determination, type matching, segment location to confidence level assessment. It solves the problems of traditional methods that are difficult to distinguish between grid disturbances and fault characteristics, false alarms and missed alarms caused by benchmark deviation, and lack of credibility assessment of identification results.

[0012] In summary, the technical solution of this application realizes the quantitative modeling of the pollution effect of power grid quality disturbances, and separates the disturbance component from the mixed signal and compensates for load characteristics. This enables multi-node high-frequency synchronous acquisition of distribution line open circuit faults, power grid disturbance identification and load characteristic pollution compensation, open circuit fault identification and confidence assessment. Through time synchronous acquisition, Fourier transform and envelope analysis, pollutant effect mathematical model, deviation vector and topology positioning, and dynamic adjustment of comprehensive pollution coefficient, it effectively improves the technical problems of the existing technology, such as the difficulty in effectively distinguishing between power grid quality disturbances and open circuit faults, the high deployment cost of contact sensors, and the high false alarm and missed alarm rate of non-intrusive monitoring methods under disturbance pollution. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 A flowchart illustrating the method for detecting loose connections and open circuits in power distribution lines based on load characteristic separation, provided in this application embodiment; Figure 2 A flowchart illustrating the identification of open circuit faults in a power distribution line based on load characteristic separation, as provided in this application embodiment. Detailed Implementation

[0015] This application provides a method for detecting loose connection and open circuit faults in distribution lines based on load characteristic separation, which solves the technical problems in the prior art where it is difficult to effectively distinguish between power grid quality disturbances and loose connection and open circuit faults, and where non-intrusive monitoring methods have high false alarm and false alarm rates under disturbance and pollution conditions.

[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. It should be noted that the numerical values ​​in the embodiments are for illustrative purposes only and do not constitute a limitation on this application.

[0017] Example 1, as shown in the appendix Figure 1 As shown, this application provides a method for detecting loose connection open circuit faults in power distribution lines based on load characteristic separation. The method includes the following steps: S100: Load current sampling data and voltage sampling data obtained by high-frequency sampling on the target distribution line; In this embodiment of the application, in scenarios where it is difficult to monitor loose connection faults in power distribution lines, waveform acquisition of the line is required to obtain load current and voltage data for fault identification. However, several problems are encountered during the acquisition process: First, the current and voltage data are difficult to correspond precisely in the time dimension, resulting in the inability to accurately correlate and analyze load characteristics with the power grid status; second, the sampling rate is insufficient to capture the rapid transient changes caused by loose connection faults, and key fault information may be lost; third, the acquisition duration is too short to reflect the evolution trend of load characteristics over time, making it difficult to establish a stable and reliable benchmark.

[0018] Step S100 in the method provided in this application embodiment includes: Sampling devices are deployed on the low-voltage side of the transformer and at the end nodes of the line to achieve synchronous acquisition of current and voltage with time synchronization accuracy meeting preset synchronization requirements. The sampling frequency is set to a preset frequency to meet high-frequency sampling requirements, and waveform data of multiple consecutive power frequency cycles are collected as the load current sampling data and voltage sampling data. A detailed explanation follows: In this embodiment, the low-voltage side of the transformer refers to the output terminal of the secondary side of the distribution transformer, typically with a voltage level of 220V / 380V, and is the power supply starting point of the user-side line. The line end node refers to the farthest branch point or load connection point of the distribution line, serving as the final reference position for the line's electrical characteristics. Time synchronization accuracy refers to the upper limit of the consistency deviation between the sampling devices at both ends in absolute time, for example, a deviation not exceeding 10μs, used to ensure time alignment of voltage and current waveforms during phase analysis. High-frequency sampling refers to a sampling method with a sampling frequency significantly higher than the power frequency, for example, a sampling frequency set to 12.8kHz, meaning 256 sampling points are collected per 50Hz power frequency cycle. Waveform data refers to a sequence of instantaneous voltage or current sampling values ​​arranged in chronological order, continuously covering multiple complete power frequency cycles, used to extract steady-state and transient electrical characteristics.

[0019] In this step, firstly, in order to achieve a precise correspondence between current and voltage in the time dimension, sampling devices need to be deployed on the low-voltage side of the transformer and the end node of the line. The sampling time deviation of the two devices is ensured to meet the preset synchronization requirements through a time synchronization mechanism, so that the voltage and current waveforms collected at the same time can be used for correlation analysis, eliminating the load characteristic calculation error caused by time misalignment. For example, the two devices achieve time synchronization through GPS time synchronization, and the synchronization deviation is controlled within 10μs.

[0020] Furthermore, in order to fully capture the high-frequency transient characteristics generated by the open circuit fault, the sampling frequency needs to be set to meet the high-frequency sampling requirements. This ensures that the sampling rate is sufficient to record the rapid changes in the circuit current, thereby obtaining the high-frequency components and transient spike information retained in the waveform. This solves the problem of circuit characteristics being smoothed or lost at low sampling rates. For example, if the sampling frequency is set to 12.8kHz, harmonic components up to 6.4kHz can be captured.

[0021] Finally, in order to reflect the changing trend of load characteristics over time, waveform data of multiple power frequency cycles need to be continuously collected to obtain voltage and current waveform sequences with sufficient time span, providing a statistically complete sample for subsequent disturbance identification and benchmark comparison. For example, waveform data of 10 power frequency cycles totaling 200ms can be continuously collected, containing 2560 sampling points.

[0022] For example, taking a 380V three-phase distribution line as an example, sampling devices are deployed at the low-voltage side of the transformer and at the end node of the line. Time synchronization is achieved through GPS timing, with a synchronization accuracy of 10μs. The sampling frequency is set to 12.8kHz, that is, 256 points are collected every 50Hz power frequency cycle. Continuous sampling is performed for 10 power frequency cycles, totaling 200ms, obtaining the initial values ​​of the A-phase voltage sampling data sequence as 311.1V, 309.8V, 308.2V, 306.9V…, and the initial values ​​of the A-phase current sampling data sequence as 15.2A, 14.8A, 14.5A, 14.6A…. Phases B and C are collected synchronously, and all three-phase data are aligned according to timestamps. A total of 2560 sets of voltage and current sampling values ​​are obtained, which serve as the load current sampling data and voltage sampling data output in step S100.

[0023] In summary, this step involves deploying sampling devices on the low-voltage side of the transformer and at the end nodes of the line to continuously acquire current and voltage waveform data for multiple power frequency cycles using a time-synchronized approach and high sampling frequency. This addresses the problems in existing technologies: firstly, it achieves synchronous acquisition of current and voltage data, eliminating correlation analysis biases caused by time misalignment; secondly, a sufficient sampling rate fully captures the transient characteristics of open-circuit faults; and thirdly, the multi-cycle continuous waveforms provide a complete temporal data foundation for subsequent disturbance identification, feature extraction, and benchmark comparison.

[0024] S200: Based on the voltage sampling data, perform power grid quality disturbance identification to obtain current disturbance characteristic data, wherein the current disturbance characteristic data includes disturbance type and disturbance intensity parameters; In this embodiment, after obtaining voltage waveform data, feature extraction is required to determine whether there are quality disturbances in the current power grid. However, several problems arise during the analysis: first, there is a lack of quantitative indicators for the degree of voltage waveform distortion, making it impossible to determine whether harmonic pollution exists; second, there is a lack of methods for extracting transient change characteristics of voltage amplitude, making it impossible to identify the occurrence and severity of sag events; and third, there is a lack of analytical capabilities for the rapid fluctuation characteristics of voltage amplitude, making it impossible to detect the existence of flicker phenomena. These problems result in the inability to accurately identify the type and intensity of power grid quality disturbances, thus failing to provide disturbance parameter input for subsequent load characteristic compensation.

[0025] Step S200 in the method provided in this application embodiment includes: A Fast Fourier Transform (FFT) is performed on the voltage sampling data to calculate the total harmonic distortion (THD). If the THD exceeds a preset THD threshold, the disturbance type is labeled as harmonic distortion. The fundamental amplitude envelope is extracted to calculate the voltage sag depth. If the sag depth exceeds a preset sag depth threshold, the disturbance type is labeled as voltage sag. The fundamental amplitude fluctuation component is extracted to calculate the voltage flicker intensity. If the flicker intensity exceeds a preset flicker intensity threshold, the disturbance type is labeled as voltage flicker. Detailed explanation follows: In this embodiment, the total harmonic distortion rate of voltage refers to the ratio of the square root of the sum of the squares of the effective values ​​of all harmonic components (excluding the fundamental wave) in the voltage waveform to the effective value of the fundamental wave, as expressed in the formula: Where h is the harmonic order, The effective value of the h-th harmonic voltage. This represents the effective value of the fundamental voltage, in volts (V). The fundamental amplitude envelope is the trajectory curve formed by connecting the amplitude points of each cycle in chronological order after extracting the fundamental voltage amplitude for each power frequency cycle of the voltage waveform; it reflects the slow change trend of the voltage amplitude. The voltage sag depth is the ratio of the drop in fundamental voltage amplitude from its rated value to the rated value, expressed by the formula: 100%, of which This is the rated voltage amplitude. The minimum amplitude of the fundamental voltage during the sag is expressed in volts (V). Voltage flicker intensity refers to the statistical value of the fluctuation amplitude of the fundamental voltage amplitude component per unit time, which is obtained by calculating the short-time flicker severity index Pst after bandpass filtering the fundamental voltage amplitude envelope.

[0026] In this step, firstly, to determine whether harmonic pollution exists in the voltage waveform, a Fast Fourier Transform (FFT) is performed on the voltage sampling data to calculate the total harmonic distortion (THD). When the THD exceeds a preset threshold, the disturbance type for that period is marked as a harmonic distortion type, thus obtaining a quantitative assessment result of harmonic pollution and solving the problem of automatic detection of harmonic distortion. For example, if the preset harmonic distortion threshold is 5%, the currently calculated THD... If it is 7.2%, it is marked as harmonic distortion type.

[0027] Furthermore, in order to identify the occurrence and severity of voltage sag events, it is necessary to extract the fundamental amplitude envelope from the voltage waveform, find the minimum value of the fundamental amplitude along the envelope, and calculate the voltage sag depth. When the sag depth is greater than a preset threshold, the disturbance type of that period is marked as a voltage sag, thus obtaining the identification result and depth quantification value of the sag event, solving the problem that voltage sags cannot be automatically captured. For example, if the preset sag depth threshold is 10%, and the current fundamental amplitude drops from 311V to 240V, the sag depth is 22.8%, then it is marked as a voltage sag.

[0028] Finally, to detect voltage flicker, the fluctuation component needs to be extracted from the fundamental amplitude envelope, and the voltage flicker intensity needs to be calculated. When the flicker intensity is greater than a preset threshold, the disturbance type for that period is marked as voltage flicker. This provides a quantitative index of rapid voltage amplitude fluctuations, solving the problem of flicker not being automatically detected. For example, if the preset flicker intensity threshold Pst is 1.0, and the currently calculated Pst is 1.5, then it is marked as voltage flicker. The above three disturbance markings can coexist, meaning that a composite disturbance of multiple disturbance types can be marked in the same period.

[0029] For example, taking a 380V three-phase distribution line as an example, the A-phase voltage waveform data of 10 power frequency cycles totaling 200ms are analyzed. First, a Fast Fourier Transform is performed to calculate the fundamental frequency as 50Hz, the effective value of the fundamental voltage as 220.0V, the square root of the sum of the squares of the effective values ​​of the 2nd to 50th harmonics as 15.8V, and the total harmonic distortion rate. The voltage sag is 7.2%, exceeding the preset harmonic distortion threshold of 5%, and the disturbance type is marked as harmonic distortion. The fundamental amplitude envelope is extracted; within 200ms, the fundamental amplitude drops from its rated value of 311V to a minimum of 248V, with a voltage sag depth of 20.3%, exceeding the preset sag depth threshold of 10%, and the disturbance type is marked as voltage sag. Bandpass filtering is applied to the fundamental amplitude envelope to extract the fluctuation component. The short-time flicker severity Pst is calculated to be 0.8, less than the preset flicker intensity threshold of 1.0, and is not marked as voltage flicker. The final current disturbance characteristic data are: disturbance types include harmonic distortion and voltage sag, and the disturbance intensity parameter is... =7.2%, Descent depth =20.3%.

[0030] In summary, this step calculates the total harmonic distortion (THD), voltage sag depth, and voltage flicker intensity by performing Fast Fourier Transform (FFT) and envelope analysis on the voltage sampling data, and compares these values ​​with preset thresholds to determine the type and intensity of the disturbance. This addresses the following problems in existing technologies: first, it enables quantitative detection of harmonic pollution through harmonic distortion rate calculation; second, it enables automatic identification of voltage sag events through envelope analysis; and third, it enables the detection of voltage flicker phenomena through amplitude fluctuation component extraction. This provides accurate disturbance type and intensity parameters for subsequent queries of mathematical models of pollutant effects.

[0031] S300: Based on the current disturbance characteristic data, query the pre-constructed mathematical model of pollutant effects to obtain the predicted value of the imprint offset; In this embodiment, after obtaining the type and intensity parameters of the current power grid quality disturbance, a mapping relationship between the disturbance parameters and the load characteristic offset needs to be established in advance to compensate for the pollution of load characteristics by the disturbance. However, several problems are encountered in establishing this mapping relationship: First, there is a lack of quantitative data on the deviation of load characteristics from the true value under disturbance conditions, making it impossible to determine the influence of different disturbance types and intensities on load imprints; second, there is a lack of predictive means from disturbance parameters to characteristic offsets, making it impossible to quickly calculate the load characteristic offset based on real-time disturbance information.

[0032] Step S300 of the method provided in this application embodiment includes: collecting first load characteristic data of a typical load type under ideal power supply conditions; collecting second load characteristic data of the typical load type under various preset disturbance conditions; calculating the difference between the second load characteristic data and the first load characteristic data to obtain the measured value of the imprint offset; using the disturbance type and disturbance intensity parameters of the preset disturbance conditions as input, and the measured value of the imprint offset as supervision, training a regression model to obtain the mathematical model of the pollutant effect. A detailed explanation follows: In this application embodiment, typical load type refers to common electrical equipment categories that account for a high proportion in the distribution network, including resistive loads such as electric water heaters, inductive loads such as asynchronous motors, and rectifier loads such as LED lighting power supplies. Ideal power supply conditions refer to a power supply environment where the voltage waveform is a standard sine wave, the amplitude is the rated value, the frequency is 50Hz and there are no harmonic components, sags, or flicker. Imprint offset refers to the difference vector between the measured load characteristics under grid disturbance conditions and the reference load characteristics under ideal conditions, including active power offset, reactive power offset, total harmonic distortion rate (THD) offset, and current waveform coefficient offset. Regression model refers to a supervised learning model that establishes the mapping relationship between input variables and continuous output variables. In this application embodiment, it is used to predict each component of the imprint offset based on the disturbance type and disturbance intensity.

[0033] In this step, firstly, in order to obtain quantitative data on the impact of different disturbance conditions on load characteristics, it is necessary to collect the first load characteristic data of typical load types under ideal power supply conditions as a benchmark, including recording active power under a standard sinusoidal power supply environment. reactive power Total harmonic distortion of current and current waveform coefficient This allows us to obtain a health baseline imprint of the load.

[0034] Furthermore, in order to obtain the influence law of disturbance on load characteristics, it is necessary to apply multiple preset disturbance conditions under the same typical load and collect second load characteristic data. The preset disturbance conditions include single disturbances and composite disturbances. Single disturbances are set according to the combination of disturbance type and intensity gradient, such as under harmonic distortion conditions. There are four gradients: 3%, 5%, 7%, and 10%. Under voltage sag conditions, there are three gradients: 10%, 20%, and 30%. Under voltage flicker conditions, there are three gradients: Pst, 0.5, 1.0, and 2.0.

[0035] Furthermore, to quantify the load characteristic deviation caused by the disturbance, it is necessary to calculate the difference between the second load characteristic data and the first load characteristic data under the same load and operating conditions, obtain the measured value of the imprint offset, and thus obtain the offset sample dataset under different disturbance conditions, for example, in Under 7% harmonic distortion conditions, the active power offset is +12W, the reactive power offset is +8Var, and the current... The offset is +3.5%, and the waveform coefficient offset is +0.02.

[0036] Finally, in order to establish the ability to predict imprint offsets from perturbation parameters, a regression model needs to be trained with perturbation type and perturbation intensity parameters as inputs and measured imprint offsets as supervised labels. A large number of offset samples are input into the model for learning to obtain a mathematical model of pollutant effects. This model can quickly output predicted imprint offsets based on real-time perturbation feature data.

[0037] For example, the regression model is built as follows: For example, taking a 380V three-phase power distribution line as an example, the core formula of the mathematical model of pollutant effects is a multiple linear regression model: =W·X+b, where X is the input vector, which contains the perturbation type code and the perturbation strength parameter. The perturbation type uses one-hot encoding, and the perturbation strength parameter is a continuous value. The output vector contains the active power offset. P, reactive power offset Q. Current total harmonic distortion offset and current waveform coefficient offset FF, with units of W, Var, % and dimensionless respectively; W is the weight matrix, and b is the bias vector. Training uses the Adam optimizer with a learning rate of 0.001 and mean squared error as the loss function. The training dataset is divided into training, validation, and test sets in an 8:1:1 ratio, with a batch size of 32 and 500 iterations. An early stopping mechanism is triggered when the validation set loss does not decrease for 20 consecutive iterations. The number of model parameters is related to the number of input and output features. Taking 3 perturbation types and a 4-dimensional output as an example, the model has a total of 24 trainable parameters.

[0038] For example, taking a 380V three-phase distribution line as an example, the disturbance type is harmonic distortion type and voltage sag type, and the disturbance intensity parameter is... =7.2%, Sag depth =20.3%. The input vector X = [harmonic distortion type encoding 1,0,0; voltage sag type encoding 0,1,0;] is used. =7.2%; Descent depth=20.3%] Input the trained mathematical model of pollutant effects, and the model outputs the predicted imprint offset value. P = +15.3W Q = +10.8Var =+4.2%, FF = +0.03. This offset prediction indicates that under the current combined disturbance conditions, the active power, reactive power, current harmonic distortion rate, and waveform coefficient of the load are all higher than the actual values, and compensation and correction are required in subsequent steps.

[0039] In summary, this step pre-collects characteristic data of typical loads under ideal and various disturbance conditions, calculates measured imprint offset values, and trains a regression model using disturbance parameters as input and offset values ​​as output, thus constructing a mathematical model of pollutant effects. This addresses the following problems in existing technologies: First, comparative experiments obtained quantitative data on imprint offsets under different disturbance conditions, clarifying the impact of disturbances on load imprints; second, regression modeling established a predictive capability from disturbance parameters to characteristic offsets, enabling the model to retrieve predicted imprint offset values ​​based on current disturbance characteristic data during actual operation, providing a quantitative basis for subsequent load characteristic pollution compensation.

[0040] S400: Based on the predicted imprint offset value and the sampled load current data, perform load characteristic contamination compensation to obtain compensated clean load characteristic data; In this embodiment, after obtaining the predicted imprint offset value, the measured load characteristic data needs to be corrected to restore the true electrical characteristics of the load. However, several problems are encountered in the correction process: First, there is a lack of systematic feature extraction methods from the original current sampling data, making it impossible to obtain the structured components of the measured load imprint; second, there is a lack of calculation methods to correspondingly reduce the offset caused by disturbances with the measured characteristics, making it impossible to separate the pure load characteristics from the contaminated measured data, resulting in the load characteristics used for subsequent fault identification still containing disturbance contamination, affecting the accuracy of identification.

[0041] Step S400 in the method provided in this application embodiment includes: The measured load imprint is extracted from the load current sampling data. The measured load imprint includes measured active power, measured reactive power, measured total harmonic distortion (THD) of the current, and measured current waveform coefficient. The predicted offset of each component is obtained from the predicted offset value. The difference between the corresponding component of the measured load imprint and the predicted offset is calculated, and this difference is used as the component of the pure load imprint. Detailed explanation follows: In this embodiment, the measured load imprint refers to a multi-dimensional electrical characteristic vector directly calculated from load current sampling data, reflecting the electrical behavior characteristics of the load under the current operating state. The measured active power refers to the average power consumed by the load within one power frequency cycle, calculated using the following formula: Where N is the number of sampling points in one cycle, and u(k) and i(k) are the instantaneous voltage and current values ​​at the k-th sampling point, respectively, in W. Measured reactive power refers to the ineffective power exchanged between the load and the power source within one power frequency cycle, calculated using the phase difference between voltage and current, as shown in the formula: ,in and These represent the effective values ​​of voltage and current, respectively, with φ being the power factor angle, measured in Var. The current waveform coefficient is the ratio of the effective value of the current waveform to the rectified average value, expressed by the formula: ,in This is the effective value of the current. The absolute value of the current is the average value over one period, which is dimensionless. The waveform coefficient is used to characterize the degree to which the current waveform deviates from a sine wave.

[0042] In this step, firstly, in order to obtain structured load characteristics from the raw current sampling data, multidimensional feature extraction needs to be performed on the load current sampling data. This includes calculating the measured active power, measured reactive power, measured total harmonic distortion rate of the measured current, and measured current waveform coefficient to form a measured load imprint vector. This provides a complete electrical characteristic description of the load under pollution conditions, solving the problem that the raw waveform data cannot be directly used for deviation analysis. For example, the extracted measured active power is 991W, the measured reactive power is 220Var, the measured total harmonic distortion rate of the measured current is 12.5%, and the measured current waveform coefficient is 1.18.

[0043] Furthermore, in order to map the offset caused by the disturbance to each component of the measured feature, the active power offset needs to be extracted from the imprint offset prediction values ​​obtained in the preceding steps. P, reactive power offset Q. Current total harmonic distortion offset and current waveform coefficient offset FF provides the predicted offset for each component, offering correction parameters for component reduction.

[0044] Finally, in order to obtain the pure load characteristics unaffected by grid disturbances, it is necessary to calculate the difference between each component of the measured load imprint and the corresponding predicted offset. The differences of each component are combined into a pure load imprint vector, which yields the compensated true load electrical characteristics. This solves the problem that the measured characteristics deviate from the true value due to disturbances. For example, the pure active power is 991W-15W=976W, the pure reactive power is 220Var-11Var=209Var, the pure current total harmonic distortion rate is 12.5%-4.2%=8.3%, and the pure current waveform coefficient is 1.18-0.03=1.15.

[0045] For example, taking a 380V three-phase distribution line as an example, the measured load imprint is extracted from the collected A-phase load current sampling data. Within 10 power frequency cycles, the effective value of the A-phase voltage is 215.0V, the effective value of the current is 4.72A, and the power factor angle is 12.5°. The calculated measured active power is 991W, and the measured reactive power is 220Var. Performing a Fast Fourier Transform on the current waveform yields an effective value of 4.65A for the fundamental current and a sum of squares of the effective values ​​of the 2nd to 50th harmonics of 0.34A. 2 The calculated total harmonic distortion (THD) rate of the measured current is 12.5%. The average absolute value of the current over one cycle is 4.00 A, and the calculated waveform coefficient of the measured current is 4.72 / 4.00 = 1.18. The predicted imprint offset value obtained from the preceding steps is... P = +15.3W Q = +10.8Var =+4.2%, FF = +0.03. Calculate the components of the pure load signature: Pure active power is 991 - 15.3 ≈ 976 W, pure reactive power is 220 - 10.8 = 209.2 Var, pure current total harmonic distortion rate is 12.5% ​​- 4.2% = 8.3%, and pure current waveform coefficient is 1.18 - 0.03 = 1.15.

[0046] In summary, this step extracts the multidimensional components of the measured load imprint from the load current sampling data, obtains the predicted offset of the corresponding component from the imprint offset prediction value, and completes pollution compensation by subtracting corresponding components to obtain clean load characteristic data. This addresses the problems in existing technologies: first, it constructs a structured measured load imprint vector by extracting active power, reactive power, total harmonic distortion rate of current, and current waveform coefficients; second, it achieves quantitative compensation for disturbance pollution through corresponding component subtraction, providing clean load characteristics unaffected by grid disturbances for subsequent fault identification.

[0047] S500: Based on the pure load characteristic data and the voltage sampling data, perform the identification of the loose connection fault and obtain the fault identification result.

[0048] In this embodiment, after obtaining the pure load characteristic data, to determine whether there is a loose connection or open circuit fault in the distribution line and to pinpoint the fault location, the pure characteristics need to be compared and analyzed with the health status benchmark. However, during the analysis process, there is a lack of systematic methods for quantifying the deviation between the pure characteristics and the benchmark, determining anomalies, differentiating fault types, and tracing fault locations. Furthermore, if the benchmark data remains fixed for a long period, it will not be able to adapt to the slow changes in load due to seasonality and aging, leading to benchmark deviations and misjudgments. In addition, when grid disturbances are severe, even after compensation, disturbance components may still remain in the pure characteristics. In this case, the reliability of the fault identification conclusion lacks assessment methods, making it difficult for maintenance personnel to judge the reliability of the identification results. The process flow for this step is as follows: Figure 2 .

[0049] Step S500 in the method provided in this application embodiment includes: The pure load characteristic data is compared with pre-stored benchmark load characteristic data to calculate a deviation vector. When the absolute value of any deviation in the deviation vector is greater than the corresponding deviation threshold, it is determined that there is a loose connection fault. The fault type is matched according to the sign and relative magnitude of each component in the deviation vector. Combined with the topology connection matrix of the target distribution line, topology backtracking is performed to obtain the line section where the fault is located. Detailed explanation is as follows: In this embodiment, the reference load characteristic data refers to the load characteristic reference value vector accumulated and calculated using the moving average filtering method during normal operation periods when the distribution line is free from power grid quality disturbances and fault characteristics. This vector includes an active power reference. Reactive power benchmark Current Total Harmonic Distortion Rate Standard and current waveform coefficient reference The deviation vector refers to the vector of differences between each component of the pure load characteristic and the corresponding component of the benchmark load characteristic, expressed by the formula: The units are W, Var, % and dimensionless, respectively. The topology connection matrix is ​​an adjacency matrix that describes the connection relationships between nodes in a power distribution line. A value of 1 in the matrix element indicates that the two nodes in the corresponding row and column are directly connected, and a value of 0 indicates that they are not directly connected. It is used to trace the propagation path of fault signals.

[0050] In this step, firstly, to quantify the difference between clean load characteristics and healthy state, the clean load characteristic data needs to be compared item by item with pre-stored baseline load characteristic data. The deviation value of each component is calculated, forming a deviation vector. This provides a multi-dimensional measure of characteristic deviation, addressing the problem that a single indicator cannot comprehensively reflect load anomalies. For example, the clean load characteristic is... =976W =209.2Var =8.3%, The baseline feature is =980W =180Var =5.0%, =1.10, then the deviation vector is [-4W, +29.2Var, +3.3%, +0.05].

[0051] Furthermore, in order to determine whether there are characteristics of a loose connection or open circuit fault, the absolute value of each deviation in the deviation vector needs to be compared with the corresponding preset deviation threshold one by one. When the absolute value of any deviation is greater than the corresponding threshold, it is determined that there are fault characteristics. This can achieve sensitive capture of weak fault signals. For example, the preset reactive power deviation threshold is 15Var. The current reactive power deviation is 29.2Var, which exceeds the threshold, and the THDI deviation is 3.3%, which exceeds the 2% threshold. Therefore, it is determined that there are characteristics of a loose connection or open circuit fault.

[0052] Finally, in order to identify the fault type and locate the fault location, it is necessary to match the preset fault type pattern library according to the sign and relative size of each component in the deviation vector, and call the topology connection matrix of the target power distribution line to trace back step by step along the direction of fault signal propagation to obtain the line section where the fault is located. For example, when the active power deviation is negative, the reactive power deviation is positive and the current waveform coefficient deviation is positive, it is matched as a series loose connection fault. The topology backtracking locates the fault in the section between the 3rd and 4th nodes.

[0053] Step S500 in the method provided in this application embodiment further includes: During periods deemed free of power grid quality disturbances and fault characteristics, clean load characteristic data for multiple consecutive periods are collected. A moving average filter is then applied to this clean load characteristic data to update the baseline load characteristic data. A detailed explanation follows: In this embodiment of the application, the moving average filtering refers to a data processing method that calculates a weighted or arithmetic mean of the pure load characteristic data of multiple consecutive periods in chronological order to smooth random fluctuations and extract steady-state trends. The window width can be 10 power frequency cycles.

[0054] In this step, in order to adapt to the slow changes in load characteristics caused by seasonality and aging, it is necessary to collect clean load characteristic data for multiple consecutive periods during the time when the system determines that there is no power grid quality disturbance and no fault characteristics. After performing moving average filtering on the collected multi-period data, the baseline load characteristic data is replaced or corrected. This yields an adaptive baseline that tracks the slow trend of load change, solving the problem of misjudgment caused by the fixed baseline gradually deviating from the true value due to load aging. For example, if clean active power data for 10 consecutive periods are collected during the period without disturbance or fault, the moving average is 998W, and the baseline is updated from 980W to 998W.

[0055] Step S500 in the method provided in this application embodiment further includes: Calculate the sum of squares of the offsets of each component in the predicted imprint offset to obtain a comprehensive contamination coefficient. When the comprehensive contamination coefficient is greater than a first contamination threshold, the confidence level of the fault identification result is reduced by a first proportion. When the comprehensive contamination coefficient is greater than a second contamination threshold and less than or equal to the first contamination threshold, the confidence level of the fault identification result is reduced by a second proportion, wherein the second contamination threshold is less than the first contamination threshold, and the second proportion is less than the first proportion. Output the reduced confidence level and the fault identification result. Detailed explanation is as follows: In this embodiment of the application, the comprehensive pollution coefficient refers to the arithmetic square root of the sum of the squares of the offsets of each component in the imprint offset prediction value, and the formula is: The unit is a comprehensive dimensionless index for each component unit, used to assess the degree of influence of residual disturbances in the pure characteristics after compensation. Confidence level refers to the percentage of credibility of the fault identification results, ranging from 0 to 100%, where 100% indicates complete confidence and 0% indicates complete unconfidence.

[0056] In this step, firstly, to quantify the potential residual perturbation impact in the compensated pure features, it is necessary to calculate the sum of squares of the offsets of each component in the predicted imprint offset value to obtain the comprehensive contamination coefficient. This coefficient provides a comprehensive evaluation index of the degree of influence of the current perturbation on the identification results, for example... =15.3W Q=10.8Var =4.2%, FF=0.03, the calculated comprehensive pollution coefficient is: 19.2.

[0057] Furthermore, in order to dynamically adjust the credibility of the identification results according to the degree of pollution, the comprehensive pollution coefficient needs to be compared with the preset pollution threshold. When the pollution coefficient is large, the confidence level is reduced proportionally to obtain the fault identification results with confidence level labels, thus solving the problem of unknown reliability of the identification conclusions under severe disturbances. For example, if the first pollution threshold is 30 and the first reduction ratio is 30%, the second pollution threshold is 15 and the second reduction ratio is 15%, and the current comprehensive pollution coefficient of 19.2 is between the two, the confidence level is reduced from 100% to 85% before output.

[0058] For example, taking a 380V three-phase distribution line as an example, the characteristics of a pure load are: =976W =209.2Var =8.3%, The current storage baseline load characteristics are P_ref=980W. =180Var =5.0%, =1.10. The deviation vector is calculated as [-4W, +29.2Var, +3.3%, +0.05]. The preset deviation thresholds are 15W for active power and 15Var for reactive power. 2%, FF 0.03. The reactive power deviation of 29.2 Var exceeds the 15 Var threshold, and the THDI deviation of 3.3% exceeds the 2% threshold, indicating a potential loose connection fault. The deviation vector shows negative P, positive Q, and positive FF, matching a series loose connection fault. Based on the topology matrix, the fault is located between nodes 3 and 4. The comprehensive pollution coefficient is calculated to be 19.2, between the first threshold of 30 and the second threshold of 15, with the confidence level lowered by 15% to 85%. The final fault identification result is: fault occurred, fault type is series loose connection fault, fault section is between nodes 3 and 4, confidence level 85%.

[0059] In summary, this step calculates the multi-dimensional deviation between the clean load characteristics and the baseline load characteristics, determines the fault characteristics based on deviation exceeding a threshold, matches the fault type according to the deviation pattern, and locates the fault segment by combining the topology connection matrix. The baseline data is automatically updated during periods of no disturbance and no faults, and the confidence level of the identification results is dynamically adjusted through a comprehensive pollution coefficient. This addresses the problems in existing technologies in the following ways: First, it establishes a complete fault identification chain from deviation calculation and threshold determination to type matching and topology location, achieving integrated fault detection, classification, and location. Second, the baseline adaptive update mechanism avoids baseline deviation caused by gradual load changes, reducing the false alarm rate. Third, the comprehensive pollution coefficient and confidence level output provide maintenance personnel with a reference for the reliability of the identification results, facilitating differentiated handling strategies based on confidence levels in actual engineering projects.

[0060] The embodiments of this application, through the specific implementation methods described above, achieve the following technical effects: This application proposes a method for detecting loose connection open circuit faults in distribution lines based on load characteristic separation. First, high-frequency synchronous data acquisition is performed on the target distribution line to obtain load current and voltage sampling data. Then, the type and intensity of grid quality disturbances are identified from the voltage data, and the predicted imprint offset value is obtained by querying a pollutant effect mathematical model. Pollution compensation is then applied to the measured load characteristics to obtain clean load characteristics. Next, the clean load characteristics are compared with a benchmark to calculate the deviation. When the deviation exceeds a threshold, the fault characteristics are determined and the fault type is matched. The fault section is located using the topology connection matrix. When grid disturbances are present, a comprehensive pollution coefficient is calculated to dynamically adjust the confidence level, outputting a structured identification result containing fault state, type, section, and confidence level. This application achieves online, disturbance-resistant, quantitative, and automated detection of loose connection open circuit faults in distribution lines through a feature separation mechanism of disturbance identification and load characteristic compensation.

[0061] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

Claims

1. A method for detecting loose connections and open circuit faults in distribution lines based on load characteristic separation, characterized in that, The method includes: Load current and voltage sampling data obtained through high-frequency sampling on the target power distribution line are loaded; Based on the voltage sampling data, power grid quality disturbance identification is performed to obtain current disturbance characteristic data, wherein the current disturbance characteristic data includes disturbance type and disturbance intensity parameters; Based on the current disturbance characteristic data, a pre-constructed mathematical model of pollutant effects is queried to obtain the predicted imprint offset value; Based on the predicted imprint offset and the sampled load current data, load characteristic pollution compensation is performed to obtain the compensated clean load characteristic data. Based on the pure load characteristic data and the voltage sampling data, a loose connection fault identification is performed to obtain the fault identification result.

2. The method as described in claim 1, characterized in that, Load current sampling data and voltage sampling data obtained through high-frequency sampling on the target distribution line are loaded, including: Sampling devices are deployed on the low-voltage side of the transformer in the distribution area and at the end nodes of the line to achieve synchronous acquisition of current and voltage with time synchronization accuracy meeting the preset synchronization requirements. The sampling frequency is set to a preset frequency that meets the high-frequency sampling requirements, and waveform data of multiple consecutive power frequency cycles are collected as the load current sampling data and voltage sampling data.

3. The method as described in claim 1, characterized in that, Based on the voltage sampling data, power grid quality disturbance identification is performed to obtain current disturbance characteristic data, including: Perform a Fast Fourier Transform on the voltage sampling data to calculate the total harmonic distortion rate (THD). When the THD is greater than a preset harmonic distortion threshold, the disturbance type is marked as a harmonic distortion type. Extract the fundamental amplitude envelope to calculate the voltage sag depth. When the voltage sag depth is greater than the preset sag depth threshold, the disturbance type is marked as voltage sag type. The voltage flicker intensity is calculated by extracting the fundamental amplitude fluctuation component. When the voltage flicker intensity is greater than the preset flicker intensity threshold, the disturbance type is marked as voltage flicker type.

4. The method as described in claim 1, characterized in that, The method for constructing the mathematical model of pollutant effects includes: Collect first load characteristic data of typical load types under ideal power supply conditions; Collect second load characteristic data of the typical load type under various preset disturbance conditions; Calculate the difference between the second load characteristic data and the first load characteristic data to obtain the measured value of the imprint offset; Using the perturbation type and intensity parameters of the preset perturbation conditions as input, and the measured value of the imprint offset as supervision, a regression model is trained to obtain the mathematical model of the pollutant effect.

5. The method as described in claim 1, characterized in that, Based on the predicted imprint offset value and the sampled load current data, load characteristic contamination compensation is performed to obtain compensated clean load characteristic data, including: The measured load imprint is extracted from the load current sampling data. The measured load imprint includes the measured active power, measured reactive power, measured total harmonic distortion rate of the measured current, and measured current waveform coefficient. Obtain the predicted offset of each component from the predicted imprint offset value; Calculate the difference between the corresponding component of the measured load imprint and the predicted offset, and use the difference as each component of the pure load imprint.

6. The method as described in claim 1, characterized in that, Based on the pure load characteristic data and the voltage sampling data, a virtual connection open circuit fault identification is performed to obtain the fault identification result, including: The pure load characteristic data is compared with the pre-stored benchmark load characteristic data, and the deviation vector is calculated. When the absolute value of any deviation in the deviation vector is greater than the corresponding deviation threshold, it is determined that there is a loose connection fault feature. The fault type is matched based on the sign and relative magnitude of each component in the deviation vector; By combining the topology connection matrix of the target power distribution line, topology backtracking is performed to locate the line section where the fault is located.

7. The method as described in claim 6, characterized in that, The method for updating the baseline load characteristic data includes: During periods when there are no power grid quality disturbances and no fault characteristics, collect clean load characteristic data for multiple consecutive cycles. The baseline load characteristic data is updated by performing a moving average filter on the pure load characteristic data of multiple consecutive periods.

8. The method as described in claim 1, characterized in that, Also includes: The sum of the squares of the offsets of each component in the predicted imprint offset is calculated as the comprehensive pollution coefficient.

9. The method as described in claim 8, characterized in that, Also includes: When the comprehensive pollution coefficient is greater than the first pollution threshold, the confidence level of the fault identification result is reduced by a first proportion; When the comprehensive pollution coefficient is greater than the second pollution threshold and less than or equal to the first pollution threshold, the confidence level of the fault identification result is reduced by a second ratio, wherein the second pollution threshold is less than the first pollution threshold and the second ratio is less than the first ratio; The adjusted confidence level is output along with the fault identification result.

Images