A cable branch box online monitoring method and system

Abstract

The application discloses a kind of cable branch box online monitoring method and system, belong to cable branch box monitoring technical field.The method includes: acquisition cable branch box inside multiple monitoring nodes real-time sensing data set, the real-time sensing data set includes three-phase current waveform sequence, zero sequence voltage fluctuation sequence and partial discharge pulse phase distribution data;According to the phase correlation characteristics of three-phase current waveform sequence and zero sequence voltage fluctuation sequence, dynamically construct the joint operation topology between the multiple monitoring nodes;Extract the coupling characteristics of the joint operation topology and partial discharge pulse phase distribution data, generate the multi-dimensional electrical characteristic map of the cable branch box;The multi-dimensional electrical characteristic map is processed by space-time entropy change analysis, and electrical abnormal event of the cable branch box is identified.The system includes processor and memory, and the processor executes the program in the memory to realize the above method.

Description

Technical Field

[0001] This invention relates to the field of cable branch box monitoring technology, specifically to an online monitoring method and system for cable branch boxes. Background Technology

[0002] As a critical node connecting main and branch lines in a power distribution network, the reliability of cable branch boxes directly impacts the continuity of power supply. The compact internal space of cable branch boxes, with densely packed multi-circuit conductors, exposes the insulation structure to the combined effects of electric, thermal, and mechanical stresses over extended periods, making them susceptible to latent faults such as partial discharge, insulation degradation, and poor contact. Conventional online monitoring methods for these faults primarily rely on threshold judgments of single electrical parameters, such as monitoring the change trends of zero-sequence voltage amplitude or partial discharge pulse counts. This approach treats each monitoring node as an isolated object, ignoring the inherent phase coupling relationships and spatial propagation characteristics between physical quantities such as three-phase current, zero-sequence voltage, and partial discharge. When surface discharge or asymmetrical decrease in inter-phase insulation strength occurs within the cable branch box, fault characteristics are dispersed across multiple monitoring nodes, manifesting in current and voltage phase shifts and partial discharge pulse phase distribution changes. Relying solely on single-point, single-parameter thresholds is insufficient to capture these correlated changes in the early stages of a fault.

[0003] Existing schemes relying on single-parameter thresholds or independent node analysis cannot characterize the phase difference fluctuations and amplitude attenuation patterns of electrical parameters in the spatial topology dimension during fault evolution. This results in insufficient early warning capabilities for the continuous accumulation of phase-to-phase short-circuit risks and the gradual formation of carbonized conductive channels on the insulation surface. Furthermore, the analysis of partial discharge signals remains at the level of statistically counting the total number of pulses or the maximum amplitude, without synchronously correlating and extracting the pulse phase distribution information with the phase characteristics of the real-time system topology. This makes it difficult to distinguish the phase shift of discharge modes caused by factors such as harmonic distortion and transient overvoltage from the original insulation defect discharge, leading to high false alarm and false negative rates in the monitoring results.

[0004] One problem this invention aims to solve is how to construct a multi-node joint operating topology that integrates current phase, voltage phase, and amplitude attenuation characteristics to reflect the spatial evolution of the current path and insulation state before a fault. Another problem is how to perform coupled feature extraction and spatiotemporal entropy change analysis on this joint operating topology and partial discharge pulse phase distribution data, and how to correlate topology parameter drift with the disorder changes in discharge pulse amplitude distribution to identify early insulation anomalies and current harmonic abrupt events. Summary of the Invention

[0005] This invention provides an online monitoring method and system for cable branch boxes. The purpose is to construct a dynamic operating topology by correlating the phase characteristics of current and voltage at multiple nodes, and to extract the coupling characteristics between the topology and the phase distribution of partial discharge for spatiotemporal entropy change processing, so as to improve the early identification capability of latent insulation faults and electrical disturbance events.

[0006] To achieve the above objectives, the present invention provides the following technical solution: The present invention provides an online monitoring method for cable branch boxes, comprising: collecting a real-time sensing data set of multiple monitoring nodes within the cable branch box, wherein the real-time sensing data set includes a three-phase current waveform sequence, a zero-sequence voltage fluctuation sequence, and partial discharge pulse phase distribution data; dynamically constructing a joint operation topology among the multiple monitoring nodes based on the phase correlation characteristics of the three-phase current waveform sequence and the zero-sequence voltage fluctuation sequence, wherein the joint operation topology includes the voltage phase angle difference matrix and the current amplitude attenuation index of each node; extracting the coupling characteristics between the joint operation topology and the partial discharge pulse phase distribution data to generate a multidimensional electrical feature map of the cable branch box, wherein the multidimensional electrical feature map includes the phase cluster center coordinates and the pulse amplitude distribution entropy; performing spatiotemporal entropy change analysis on the multidimensional electrical feature map to identify electrical abnormal events of the cable branch box, wherein the electrical abnormal events include partial discharge phase interval shifts and sudden changes in current harmonic components. This method uses correlation analysis of multi-source data and dynamic topology construction to keenly capture abnormal signs from the spatiotemporal evolution of phase and amplitude, thereby improving the detection sensitivity and real-time performance of weak electrical faults in cable branch boxes.

[0007] Preferably, the process of dynamically constructing the joint operation topology includes: performing phase-by-phase zero-crossing detection processing on the three-phase current waveform sequence, extracting the instantaneous phase angle of each phase current, and forming a phase angle time sequence of each phase current; performing peak location processing on the zero-sequence voltage fluctuation sequence, extracting the time corresponding to the zero-sequence voltage zero-crossing point as a zero-sequence voltage phase reference; performing difference calculation between the phase angle time sequence of each phase current and the zero-sequence voltage phase reference, calculating the phase difference between each phase current and the zero-sequence voltage, and generating the voltage phase angle difference matrix; and based on the voltage phase angle difference matrix, fitting the attenuation coefficient of the three-phase current amplitude of each monitoring node to generate the current amplitude attenuation index.

[0008] Further preferably, the three-phase current waveform sequence is subjected to phase-by-phase zero-crossing detection processing to extract the instantaneous phase angle of each phase current. Specifically, this includes: binarizing each phase current waveform, marking sampling points with current amplitude greater than zero as the first state, and marking sampling points with current amplitude less than or equal to zero as the second state; traversing the binarized sequence of each phase current waveform, and using the alternating boundary points between consecutive first and second state segments as candidate zero-crossing points; performing linear interpolation correction on each candidate zero-crossing point to calculate its precise zero-crossing time, and using the ratio of the interval angle between two adjacent precise zero-crossing times to the preset power frequency cycle angle as the unit increment of the instantaneous phase angle, accumulating to generate the phase angle time series. This precise zero-crossing point positioning effectively ensures the accuracy of phase angle calculation, laying the foundation for the reliable construction of the subsequent phase angle difference matrix.

[0009] Preferably, the process of extracting the coupling features of the joint operation topology and partial discharge pulse phase distribution data to generate a multidimensional electrical feature map includes: mapping each pulse in the partial discharge pulse phase distribution data to the corresponding phase interval in the voltage phase difference matrix according to its occurrence time, obtaining the pulse count and pulse amplitude sequence in each phase interval; normalizing the pulse amplitude sequence in each phase interval, calculating the pulse amplitude distribution entropy of each phase interval, and generating the distribution curve of the pulse amplitude distribution entropy as a function of phase; combining each element in the voltage phase difference matrix with the current amplitude attenuation index of the corresponding phase to generate a multidimensional node feature vector for each monitoring node; extracting the peak value from the distribution curve of the pulse amplitude distribution entropy, and using the phase coordinates corresponding to the peak value as the coordinates of the phase clustering center.

[0010] A further preferred embodiment of calculating the pulse amplitude distribution entropy for each phase interval includes: sorting the pulse amplitude sequence within each phase interval, extracting the maximum and minimum values ​​to construct an amplitude interval; dividing the pulse amplitude sequence within each phase interval into several equal amplitude portions, counting the number of pulses in each amplitude portion, and generating an amplitude-frequency histogram; calculating the proportion of the number of pulses in each amplitude portion of the amplitude-frequency histogram to the total number of pulses in that phase interval, multiplying each proportion by the natural logarithm of that proportion, summing them, and taking the negative value to obtain the pulse amplitude distribution entropy. The resulting distribution entropy can sensitively reflect the degree of disorder in the amplitude dimension of partial discharge pulses, allowing subtle changes in discharge modes caused by insulation degradation to be significantly characterized.

[0011] Preferably, the process of performing spatiotemporal entropy change analysis on the multidimensional electrical feature map to identify electrical anomalies includes: sampling the time-varying sequence of the multidimensional electrical feature map, with each sliding window covering a preset number of map time frames; calculating the Euclidean distance between the phase cluster center coordinates of every two adjacent time frames to generate a center coordinate drift sequence; performing differential processing on the temporal change of the pulse amplitude distribution entropy within the sliding window to calculate the rate of change of entropy over time and generate an entropy change rate sequence; aligning and fusing the center coordinate drift sequence and the entropy change rate sequence to generate a comprehensive anomaly index sequence, which is a weighted sum of the center coordinate drift sequence and the entropy change rate sequence; taking the time corresponding to the element in the comprehensive anomaly index sequence that exceeds a preset threshold as the occurrence time of the electrical anomaly event, and extracting the phase cluster center coordinates at that time as the offset of the partial discharge phase interval shift, while simultaneously extracting the entropy change rate at that time as the abrupt change amplitude of the current harmonic component change.

[0012] A further preferred embodiment of the process of aligning and fusing the center coordinate drift sequence and the entropy change rate sequence to generate a comprehensive anomaly index sequence includes: normalizing the center coordinate drift sequence by subtracting the mean of the sequence from each element and then dividing by the standard deviation of the sequence to obtain a normalized drift sequence; absoluteifying the entropy change rate sequence by replacing each element with its absolute value to obtain an absolute value entropy change rate sequence; normalizing the absolute value entropy change rate sequence by subtracting the mean of the sequence from each element and then dividing by the standard deviation of the sequence to obtain a normalized entropy change rate sequence; and adding the elements at corresponding times in the normalized drift sequence and the normalized entropy change rate sequence to generate the comprehensive anomaly index sequence. This fusion method enables phase shift and entropy change, two types of anomaly features, to mutually reinforce each other at a unified scale, significantly improving the robustness and accuracy of anomaly event identification.

[0013] As a preferred embodiment of the present invention, the method further includes noise filtering of the partial discharge pulse phase distribution data: extracting time-domain waveform features for each pulse in the partial discharge pulse phase distribution data, and calculating three time-domain feature parameters: pulse rise time, pulse fall time, and pulse width; comparing the pulse rise time, pulse fall time, and pulse width with standard pulse parameters in a preset partial discharge typical pulse database one by one, and calculating the Euclidean distance between the feature parameters of each pulse and each standard pulse parameter; using the minimum Euclidean distance corresponding to each pulse as the matching residual of the pulse, retaining pulses with matching residuals less than a preset residual threshold, and marking pulses with matching residuals greater than or equal to the preset residual threshold as noise pulses and removing them from the partial discharge pulse phase distribution data. Preferably, the preset residual threshold is 0.5 times the standard deviation of all standard pulse feature parameters in the partial discharge typical pulse database. This noise filtering method effectively suppresses the contamination of discharge pulse data by on-site electromagnetic interference, ensuring the purity and reliability of subsequent feature extraction and anomaly identification.

[0014] This invention also provides an online monitoring system for cable branch boxes, including a processor and a memory. The memory is connected to the processor and is used to store programs, instructions, or code. The processor executes the programs, instructions, or code in the memory to implement the above-described online monitoring method for cable branch boxes. This system can perform online multi-source electrical quantity acquisition, dynamic topology construction, coupled feature extraction, and spatiotemporal-entropy joint analysis. It continuously senses the operating status of cable branch boxes and provides early warnings of anomalies without power interruption. It features simple deployment, rapid response, and high accuracy.

[0015] The technical effects and advantages provided by the present invention in the above technical solution are as follows: By extracting the difference between the instantaneous phase angle of the three-phase current waveform sequence and the zero-sequence voltage phase reference, a voltage phase angle difference matrix is ​​constructed. The attenuation coefficient of the three-phase current amplitude at each node is then fitted to generate a current amplitude attenuation index, thus forming a multi-node joint operation topology that incorporates phase angle difference and amplitude attenuation characteristics. This topology is generated not by a fixed network parameter model, but dynamically adjusted according to the real-time waveform phase correlation. This ensures that even under load fluctuations and changes in operating conditions, the topology can accurately reflect the differences in electrical connection strength between conductors at each node caused by changes in insulation resistance, contact surface oxidation, or partial discharge carbonization. When a weak partial discharge or insulation dampness occurs in a certain area, the phase difference between nodes in that area and other nodes will exhibit continuous small drifts, and the corresponding current amplitude attenuation coefficient will show non-stationary fluctuations. This correlation change can be directly characterized by the joint operation topology, thus providing spatial fault source pointing information for subsequent feature analysis and overcoming the deficiency of single-node independent analysis in capturing changes in the spatial propagation path before a fault.

[0016] The joint operation topology and partial discharge pulse phase distribution data are coupled to generate a multidimensional electrical characteristic map containing phase cluster center coordinates and pulse amplitude distribution entropy. Spatiotemporal entropy variation analysis is then performed on this map. During the spatiotemporal entropy variation analysis, a sliding window is used to calculate the drift sequence of the phase cluster center coordinates and the rate of change sequence of the pulse amplitude distribution entropy. These two sequences are then aligned and fused to form a comprehensive anomaly index sequence. The drift of the phase cluster center coordinates reflects whether the concentrated phase region of the partial discharge has shifted during the power frequency cycle. This shift is usually caused by a change in the discharge mechanism or the expansion of the insulation degradation area. The pulse amplitude distribution entropy characterizes the uniformity of the discharge pulse amplitude distribution. When abrupt harmonic component changes or intermittent arcs occur, the discharge amplitude no longer concentrates in a few stable amplitude intervals but develops towards greater dispersion, leading to a sharp increase in the entropy rate. By fusing the topological space drift with the rate of change of the disorder of the discharge amplitude, it is possible to distinguish between a simple increase in discharge intensity and a dangerous discharge accompanied by topological deterioration, avoiding false alarms caused by fluctuations in a single indicator and improving the accuracy of identifying early insulation abnormalities and current harmonic mutation events. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0018] Figure 1 This is a flowchart of the online monitoring method for cable branch boxes; Figure 2This is a flowchart of dynamic topology construction based on the phase correlation characteristics of multiple monitoring nodes operating together; Figure 3 This is a flowchart of three-phase current zero-crossing detection and instantaneous phase angle extraction; Figure 4 This is a flowchart for generating a multi-dimensional electrical feature map of a cable branch box; Figure 5 This is a flowchart for identifying electrical anomalies based on spatiotemporal entropy change analysis of multidimensional electrical feature maps. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] See Figure 1 This invention provides an online monitoring method for cable branch boxes. The method includes the following steps: First, collecting real-time sensor data sets from multiple monitoring nodes within the cable branch box. These real-time sensor data sets include three-phase current waveform sequences, zero-sequence voltage fluctuation sequences, and partial discharge pulse phase distribution data. Second, dynamically constructing a joint operating topology among the multiple monitoring nodes based on the phase correlation characteristics of the three-phase current waveform sequences and the zero-sequence voltage fluctuation sequences. This joint operating topology includes the voltage phase angle difference matrix and current amplitude attenuation index of each node. Third, extracting the coupling characteristics between the joint operating topology and the partial discharge pulse phase distribution data to generate a multi-dimensional electrical feature map of the cable branch box. This multi-dimensional electrical feature map includes the phase cluster center coordinates and pulse amplitude distribution entropy. Fourth, performing spatiotemporal entropy change analysis on the multi-dimensional electrical feature map to identify electrical abnormal events in the cable branch box. These electrical abnormal events include partial discharge phase interval shifts and sudden changes in current harmonic components.

[0021] Example 1:

[0022] In specific implementation, please refer to Figure 2 Based on the phase correlation characteristics between the three-phase current waveform sequence and the zero-sequence voltage fluctuation sequence, the following method is used to dynamically construct the joint operation topology between multiple monitoring nodes.

[0023] The three-phase current waveform sequence is processed by phase-by-phase zero-crossing detection to extract the instantaneous phase angle of each phase current, forming a phase angle time sequence for each phase current. The three-phase current waveform sequence includes phase A, phase B, and phase C current waveform sequences. For the phase A current waveform sequence, iterate through any two adjacent sampling points in the phase A current waveform sequence, and the current amplitude at the current sampling point is... The current amplitude at the next sampling point is ,when and When the product of the two sampling points is less than zero, a zero-crossing point is determined to exist between them. Linear interpolation is performed using the time and amplitude of the two sampling points to calculate the precise time of this zero-crossing point in the A-phase current waveform sequence. The following formula is used to obtain:

[0024] in, The current sampling point's time. For the time of the next sampling point, The amplitude of phase A current at the current sampling point. The amplitude of phase A current at the next sampling point is recorded. The precise times of all zero-crossing points in the phase A current waveform sequence are recorded in chronological order to obtain the phase A zero-crossing time sequence. The time interval between any two adjacent precise zero-crossing times in the phase A zero-crossing time sequence is divided by the power frequency cycle length and then multiplied by 360 degrees to obtain the phase increment of the phase A current in each power frequency cycle. The phase increments of the phase A current are accumulated cycle by cycle to form the phase angle time sequence of the phase A current. For the phase B current waveform sequence and the phase C current waveform sequence, the same method is used to obtain the phase angle time sequences of the phase B current and the phase C current, respectively.

[0025] Peak location processing is performed on the zero-sequence voltage fluctuation sequence to extract the time corresponding to the zero-crossing point of the zero-sequence voltage, which is used as the phase reference for the zero-sequence voltage. The zero-sequence voltage fluctuation sequence consists of the zero-sequence voltage amplitudes of multiple sampling points. Traversing any two adjacent sampling points in the zero-sequence voltage fluctuation sequence, the zero-sequence voltage amplitude of the current sampling point is... The zero-sequence voltage amplitude at the next sampling point is ,when Less than zero and When the value is greater than zero, a zero-sequence voltage zero-crossing point is determined to exist between the two sampling points, crossing from negative to positive. The precise time of the zero-sequence voltage zero-crossing point is calculated by linear interpolation using the time and amplitude of the two sampling points. Determined by the following formula:

[0026] in, This represents the time of the current sampling point in the zero-sequence voltage fluctuation sequence. This represents the time of the next sampling point in the zero-sequence voltage fluctuation sequence. The zero-sequence voltage amplitude at the current sampling point. This is the zero-sequence voltage amplitude at the next sampling point. The precise time of each zero-crossing point of the obtained zero-sequence voltage is used as the zero-sequence voltage phase reference time.

[0027] The phase angle time series of each phase current is subtracted from the zero-sequence voltage phase reference to calculate the phase difference between each phase current and the zero-sequence voltage, generating a voltage phase angle difference matrix. For a given monitoring node, at each zero-sequence voltage phase reference time, the instantaneous phase angles of phase A current, phase B current, and phase C current are obtained from their respective phase angle time series. The instantaneous phase angle of phase A current is then subtracted from the zero-sequence voltage phase at the zero-sequence voltage phase reference time to obtain the phase difference between phase A current and the zero-sequence voltage. The zero-sequence voltage phase is set to 0 degrees at the zero-crossing point of the zero-sequence voltage. Similarly, the phase differences between phase B current and the zero-sequence voltage, as well as the phase differences between phase C current and the zero-sequence voltage, are obtained. The three-phase phase differences of multiple monitoring nodes at different zero-sequence voltage phase reference times are arranged according to the monitoring node number and phase sequence to generate a voltage phase difference matrix. Each row in the voltage phase difference matrix corresponds to a monitoring node, and each column corresponds to the phase difference of a certain phase of the monitoring node at a certain zero-sequence voltage phase reference time.

[0028] Based on the voltage phase difference matrix, the attenuation coefficient of the three-phase current amplitude at each monitoring node is fitted to generate a current amplitude attenuation index. For each monitoring node, the three-phase current amplitude at the corresponding moment is extracted from the three-phase current waveform sequence, and the root mean square value of the three-phase current amplitude is used as the representative value of the current amplitude of the monitoring node. The average value of the phase difference of phase A, phase B, and phase C of the monitoring node in the voltage phase difference matrix is ​​used as the phase offset of the monitoring node, and data pairs are formed by combining the phase offset of different monitoring nodes with the representative value of the current amplitude of the monitoring node. An exponential function is fitted to all data pairs of the monitoring nodes. The fitted function is such that the representative value of the current amplitude of the monitoring node changes exponentially with the phase offset. The resulting exponential coefficient is the current amplitude attenuation index. The current amplitude attenuation index characterizes the overall attenuation trend under the multi-node joint operation topology in the cable branch box. The change in the index value is related to the overall change in the insulation state or current conduction path in the branch box.

[0029] Example 2:

[0030] In specific implementation, please refer to Figure 3 The current waveform of each phase in the three-phase current waveform sequence is binarized. Sampling points with current amplitude greater than zero are marked as the first state, and sampling points with current amplitude less than or equal to zero are marked as the second state. The three-phase current waveform sequence consists of phase A current waveform sequences, phase B current waveform sequences, and phase C current waveform sequences. For the phase A current waveform sequence, the phase A current amplitude of each sampling point in the phase A current waveform sequence is obtained, and each phase A current amplitude is compared with zero. When the phase A current amplitude of a sampling point is greater than zero, the state of that sampling point is marked as the first state, which can be represented by the value 1; when the phase A current amplitude of a sampling point is less than or equal to zero, the state of that sampling point is marked as the second state, which can be represented by the value 0. According to the original time order of the sampling points in the phase A current waveform sequence, the marked first or second states are arranged to form the binarized sequence of phase A current waveforms. The same operation is performed on the phase B current waveform sequence to form the binarized sequence of phase B current waveforms. Perform the same operation on the C-phase current waveform sequence as on the A-phase current waveform sequence to form a binarized C-phase current waveform sequence.

[0031] The binary sequence of each phase current waveform is traversed, and the alternating boundary points between consecutive first and second state segments are used as candidate zero-crossing points. Taking the binary sequence of phase A current waveform as an example, starting from the first state value of the binary sequence of phase A current waveform, each state value is compared with its immediately following state value. When the state value changes from the first state to the second state, the alternation position of the first state segment to the second state segment is recorded, and the sampling point interval in the original phase A current waveform sequence corresponding to the alternation position is used as the first type of candidate zero-crossing point interval; when the state value changes from the second state to the first state, the alternation position of the second state segment to the first state segment is recorded, and the sampling point interval in the original phase A current waveform sequence corresponding to the alternation position is used as the second type of candidate zero-crossing point interval. The boundary sampling points of all first type and second type candidate zero-crossing point intervals are used as candidate zero-crossing points of phase A current waveform. The same operation is performed on the binary sequence of phase B current waveform to obtain candidate zero-crossing points of phase B current waveform. Perform the same operation on the binarized sequence of the C-phase current waveform to obtain candidate zero-crossing points of the C-phase current waveform.

[0032] For each candidate zero-crossing point, perform linear interpolation correction to calculate the precise zero-crossing time. For the sampling point interval containing a candidate zero-crossing point in the A-phase current waveform, obtain the time of the previous sampling point in that interval. and the time of the next sampling point and the amplitude of phase A current at the previous sampling point and the amplitude of phase A current at the next sampling point Since this interval is the interval that crosses from the positive half-cycle to the negative half-cycle or from the negative half-cycle to the positive half-cycle, and The signs are opposite. Calculate the precise zero-crossing time of this candidate zero-crossing interval in the A-phase current waveform. The formula is:

[0033] in, The time of the previous sampling point within the sampling point interval of the candidate zero-crossing point. The time of the next sampling point within the sampling point interval of the candidate zero-crossing point. This represents the amplitude of phase A current at the previous sampling point. The amplitude of phase A current at the next sampling point is used. The above linear interpolation correction process is applied to all candidate zero-crossing points in the phase A current waveform to obtain all precise zero-crossing times in the phase A current waveform. The preset power frequency cycle angle is set to 360 degrees. The time interval between two adjacent precise zero-crossing times in the phase A current waveform is taken as the zero-crossing interval duration. The zero-crossing interval duration is divided by the power frequency cycle duration and then multiplied by the preset power frequency cycle angle of 360 degrees to obtain the instantaneous phase angle unit increment of the phase A current waveform between those two adjacent precise zero-crossing times. Starting from the first precise zero-crossing time in the phase A current waveform, the initial phase angle is set to 0 degrees. Each time a zero-crossing interval duration is crossed, the instantaneous phase angle unit increment is accumulated to the previous accumulated phase angle, accumulating to generate the phase angle time series of the phase A current. The same operation is performed on the phase B and phase C current waveforms to generate the phase angle time series of the phase B current and the phase C current, respectively.

[0034] Example 3:

[0035] In specific implementation, please refer to Figure 4The partial discharge pulse phase distribution data is mapped to the corresponding phase interval in the voltage phase difference matrix according to the pulse occurrence time, resulting in a pulse count and pulse amplitude sequence within each phase interval. The partial discharge pulse phase distribution data contains records of multiple pulses, each recording including the pulse occurrence time and pulse amplitude. The voltage phase difference matrix consists of multiple rows and columns, with each column corresponding to a specific phase interval. A phase interval is a continuous angular range obtained by dividing the power frequency cycle from 0 degrees to 360 degrees according to a preset phase interval. The preset phase interval is set to a fixed angle value; for example, when 360 degrees is divided into 36 equal phase intervals, each phase interval spans 10 degrees. For a pulse in the partial discharge pulse phase distribution data, the pulse occurrence time is extracted. Based on the power frequency phase value corresponding to the pulse occurrence time, the target phase interval of the pulse is determined. The pulse amplitude is then assigned to the pulse amplitude sequence of the target phase interval, and the pulse count value of the target phase interval is incremented by 1. After performing the above mapping process on all pulses in the partial discharge pulse phase distribution data, each phase interval corresponds to a pulse count and a pulse amplitude sequence composed of all pulse amplitudes that fall into the phase interval.

[0036] The pulse amplitude sequence within each phase interval is normalized, and the pulse amplitude distribution entropy for each phase interval is calculated to generate a distribution curve of the pulse amplitude distribution entropy as a function of phase. In some embodiments, the calculation process of the pulse amplitude distribution entropy is as follows: The pulse amplitude sequence within each phase interval is sorted, and the maximum and minimum values ​​in the pulse amplitude sequence are extracted. The amplitude span is obtained by subtracting the extracted minimum value from the extracted maximum value. The amplitude interval is constructed with the extracted minimum value as the lower bound and the extracted maximum value as the upper bound. The pulse amplitude sequence within each phase interval is divided into several equal amplitude parts. The number of amplitude parts is set to a fixed value, which is determined according to the number of pulses in the pulse amplitude sequence. When the number of pulses exceeds 100, the number of amplitude parts is set to 20, and when the number of pulses does not exceed 100, the number of amplitude parts is set to 10. During the commissioning phase before system operation, the number of amplitude segments is pre-statistically determined using confirmed normal operating data from the current cable branch boxes. The appropriate number of segments is determined based on the pulse density and amplitude span. Preset thresholds and residual thresholds are also calibrated during this commissioning phase using normal operating data to ensure that each criterion parameter matches the actual electrical characteristics of the current branch box under voltage level, load level, and ambient temperature. The number of pulses within each amplitude segment is counted, generating an amplitude-frequency histogram. The horizontal axis of the amplitude-frequency histogram represents the amplitude segment number, and the vertical axis represents the number of pulses counted within each amplitude segment. The proportion of the pulse count in each amplitude segment to the total number of pulses in its phase interval is calculated. Each proportion is multiplied by its natural logarithm, summed, and the result is negative to obtain the pulse amplitude distribution entropy. The calculation formula is:

[0037] in, The total number of equal parts of the amplitude. Numbering for equal portions of the amplitude. For the first The proportion of pulses within each amplitude division to the total number of pulses in the corresponding phase interval. For proportion The natural logarithm of the phase angle. Using the center phase angle of each phase interval as the x-axis and the pulse amplitude distribution entropy calculated for each phase interval as the y-axis, the data points of each phase interval are connected in phase order to generate the distribution curve of pulse amplitude distribution entropy as a function of phase.

[0038] Each element in the voltage phase difference matrix is ​​vector-combined with the corresponding phase current amplitude attenuation index to generate a multidimensional node feature vector for each monitoring node. Each element in the voltage phase difference matrix represents the voltage phase difference of a monitoring node within a specific phase interval. The current amplitude attenuation index value for that monitoring node within the same phase interval is obtained from the current amplitude attenuation index. The voltage phase difference value and the current amplitude attenuation index value are concatenated into a two-dimensional vector, which serves as the multidimensional node feature vector for that monitoring node within that phase interval. This vector combination operation is performed on each monitoring node within each phase interval to obtain the set of multidimensional node feature vectors for each monitoring node.

[0039] Peak values ​​are extracted from the pulse amplitude distribution entropy curve, and the phase coordinates corresponding to the peak values ​​are used as the phase cluster center coordinates. All data points on the pulse amplitude distribution entropy curve are traversed. For a given data point, if its pulse amplitude distribution entropy value is greater than both the preceding and following data points, it is identified as a peak point. The phase value corresponding to the peak point is used as the phase cluster center coordinates. When multiple peak points appear on the distribution curve, their phase values ​​are used as different phase cluster center coordinates.

[0040] Example 4:

[0041] In specific implementation, please refer to Figure 5 A sliding window sampling method is used to sample the time-varying sequence of multidimensional electrical feature maps. The time-varying sequence consists of multiple time frames arranged chronologically, each corresponding to a sampling time. Each time frame contains the phase cluster center coordinates and pulse amplitude distribution entropy at that sampling time. The sliding window covers a preset number of time frames, a fixed value of 10. The sliding window moves chronologically from the start to the end of the time sequence, with each movement taking a step of one time frame. After each movement, the Euclidean distance between the phase cluster center coordinates of every two adjacent time frames within the sliding window is calculated. The phase cluster center coordinates are two-dimensional coordinates, composed of the first and second phase cluster center coordinate values. When the phase cluster center coordinates of two adjacent time frames are respectively... and At that time, the Euclidean distance between the phase cluster center coordinates of two adjacent time frames The calculation formula is:

[0042] in, The coordinates of the first phase cluster center in the previous time frame. These are the coordinates of the second phase cluster center in the previous time frame. The coordinates of the first phase cluster center in the next time frame. This represents the coordinates of the second phase cluster center in the next time frame. The Euclidean distances between the phase cluster center coordinates of all adjacent time frames within the sliding window are arranged chronologically to generate a center coordinate drift sequence. The length of the center coordinate drift sequence is equal to the preset number minus 1.

[0043] The temporal variation of pulse amplitude distribution entropy within a sliding window is differentially processed. Within a preset number of spectral time frames covered by the sliding window, each spectral time frame corresponds to a pulse amplitude distribution entropy value. The pulse amplitude distribution entropy values ​​of all spectral time frames within the sliding window are arranged into an entropy value sequence in chronological order. For each entropy value element in the entropy value sequence, the current entropy value element is subtracted from the next entropy value element, and the difference is taken as the entropy change rate corresponding to the current entropy value element. All entropy change rates within the sliding window are arranged in chronological order to generate an entropy rate sequence, the length of which is equal to a preset number minus 1.

[0044] The center coordinate drift sequence and the entropy change rate sequence are aligned and fused to generate a comprehensive anomaly index sequence, which is a weighted sum of the center coordinate drift sequence and the entropy change rate sequence. In specific implementation, the process of aligning and fusing the center coordinate drift sequence and the entropy change rate sequence to generate the comprehensive anomaly index sequence is as follows: The center coordinate drift sequence is normalized. The mean center coordinate drift of all elements in the center coordinate drift sequence is calculated, and the standard deviation of the center coordinate drift of all elements in the center coordinate drift sequence is calculated. Each element in the center coordinate drift sequence is subtracted from the mean center coordinate drift and then divided by the standard deviation of the center coordinate drift to obtain the normalized drift sequence.

[0045] The entropy change rate sequence is converted to absolute values. Each element in the sequence is replaced with its absolute value, resulting in an absolute value entropy change rate sequence. This sequence is then normalized. The mean and standard deviation of the absolute entropy change rate for all elements in the sequence are calculated. Finally, each element in the sequence is subtracted from its mean and then divided by its standard deviation to obtain the normalized entropy change rate sequence.

[0046] Add the elements at corresponding times in the normalized drift sequence and the normalized entropy change rate sequence. The element at the corresponding time in the normalized drift sequence... The element and the normalized entropy change rate sequence The elements correspond to the same time frame interval. Adding the two elements together yields the nth element in the comprehensive anomaly index sequence. Each element is added to the elements at all corresponding time points in both the normalized drift sequence and the normalized entropy change rate sequence to generate a comprehensive anomaly index sequence. The length of the comprehensive anomaly index sequence is the same as the length of the normalized drift sequence.

[0047] The time corresponding to the element in the comprehensive anomaly index sequence that exceeds a preset threshold is taken as the occurrence time of the electrical anomaly event. The preset threshold is set to a fixed value of 3.0. When the distribution of on-site monitoring data of the cable branch box deviates from a normal distribution, the actual numerical distribution of the comprehensive anomaly index sequence collected during the confirmed abnormal operation period is statistically analyzed, and the value corresponding to the upper quantile is used to replace the fixed preset threshold of 3.0. The basis for setting this fixed value is: when the elements in the comprehensive anomaly index sequence follow a normal distribution, the probability that the element value exceeds the mean plus 3 times the standard deviation is less than 0.3%, which corresponds to the low false alarm requirement of the anomaly event. Traversing each element in the comprehensive anomaly index sequence, when the value of a certain element is greater than 3.0, the time of the second of the last two time frames in the sliding window corresponding to that element is taken as the occurrence time of the electrical anomaly event. The phase cluster center coordinates in the multidimensional electrical feature map at the occurrence time of the electrical anomaly event are extracted, and the phase cluster center coordinates are used as the offset of the partial discharge phase interval. Simultaneously, the pulse amplitude distribution entropy in the multidimensional electrical feature spectrum at the moment of occurrence of the electrical anomaly is extracted, and the entropy change rate at that moment is taken as the abrupt change amplitude of the current harmonic component. The entropy change rate is the rate of change of entropy value corresponding to the moment of occurrence of the electrical anomaly within the sliding window.

[0048] Example 5:

[0049] In practical implementation, time-domain waveform features are extracted for each pulse in the partial discharge pulse phase distribution data, and three time-domain feature parameters—pulse rise time, pulse fall time, and pulse width—are calculated. The partial discharge pulse phase distribution data contains records of multiple pulses. Each pulse record contains a time-domain waveform sampling sequence of one pulse. The pulse's time-domain waveform sampling sequence consists of multiple sampling points, each containing a sampling time and pulse amplitude. For a pulse's time-domain waveform sampling sequence, all sampling points in the sequence are traversed, and the maximum pulse amplitude is extracted as the pulse peak amplitude. The sampling time corresponding to the pulse peak amplitude is recorded as the pulse peak time. The pulse peak amplitude is multiplied by a first proportionality factor (set to 0.1) to obtain the pulse upper reference amplitude, and multiplied by a second proportionality factor (set to 0.9) to obtain the pulse lower reference amplitude. In the time-domain waveform sampling sequence of the pulse, a search is performed from the pulse peak time in the direction of decreasing time to find two adjacent sampling points where the pulse amplitude rises from below the upper reference amplitude to above the upper reference amplitude. Linear interpolation is then used to obtain the first precise moment when the pulse amplitude equals the upper reference amplitude. Similarly, another search is performed from the pulse peak time in the direction of decreasing time to find two adjacent sampling points where the pulse amplitude rises from below the lower reference amplitude to above the lower reference amplitude. Linear interpolation is then used to obtain the second precise moment when the pulse amplitude equals the lower reference amplitude. The time interval obtained by subtracting the first precise moment from the second precise moment is taken as the pulse rise time. In the time-domain waveform sampling sequence of the pulse, a search is performed from the pulse peak time in the direction of increasing time to find two adjacent sampling points where the pulse amplitude drops from above the lower reference amplitude to below the lower reference amplitude. Linear interpolation is then used to obtain the third precise moment when the pulse amplitude equals the lower reference amplitude. Similarly, another search is performed from the pulse peak time in the direction of increasing time to find two adjacent sampling points where the pulse amplitude drops from above the upper reference amplitude to below the upper reference amplitude. Linear interpolation is then used to obtain the fourth precise moment when the pulse amplitude equals the upper reference amplitude. The time interval obtained by subtracting the third precise moment from the fourth precise moment is taken as the pulse fall time. The time interval obtained by subtracting the first precise moment from the fourth precise moment is taken as the pulse width. This time-domain waveform feature extraction process is performed on each pulse in the partial discharge pulse phase distribution data to obtain the pulse rise time, pulse fall time, and pulse width for each pulse.

[0050] The pulse rise time, pulse fall time, and pulse width are compared one by one with the standard pulse parameters in a pre-set database of typical partial discharge pulses. The Euclidean distance between the characteristic parameters of each pulse and each standard pulse parameter is calculated. The pre-set database of typical partial discharge pulses stores multiple standard pulse records, each containing a standard pulse rise time, a standard pulse fall time, and a standard pulse width. For a pulse in the partial discharge pulse phase distribution data, the pulse rise time, pulse fall time, and pulse width are used to form the characteristic vector of the pulse to be tested. The standard pulse rise time, standard pulse fall time, and standard pulse width of each standard pulse record in the pre-set database of typical partial discharge pulses are used to form the standard pulse characteristic vector. The Euclidean distance between the characteristic vector of the pulse to be tested and a standard pulse characteristic vector is calculated. The calculation formula is:

[0051] in, The pulse rise time is the characteristic vector of the pulse to be measured. The pulse fall time is the characteristic vector of the pulse to be measured. The pulse width is the feature vector of the pulse to be measured. The standard pulse rise time is the characteristic vector of the standard pulse. The standard pulse fall time is the characteristic vector of the standard pulse. The standard pulse width is defined in the standard pulse feature vector. The Euclidean distance between the feature vector of the pulse under test and all standard pulse feature vectors in the preset typical partial discharge pulse database is calculated. The smallest Euclidean distance among all calculated distances is selected as the matching residual for that pulse.

[0052] Pulses with matching residuals less than a preset residual threshold are retained, while pulses with matching residuals greater than or equal to the preset residual threshold are marked as noise pulses and removed from the partial discharge pulse phase distribution data. The preset residual threshold is 0.5 times the standard deviation of all standard pulse characteristic parameters in the typical partial discharge pulse database. The standard deviation of all standard pulse characteristic parameters in the typical partial discharge pulse database is calculated as follows: the rise times of all standard pulse records in the typical partial discharge pulse database are collected as a rise time sample set. The mean rise time of the rise time sample set is calculated. The rise time of each standard pulse in the rise time sample set is subtracted from the mean rise time, and the square is taken. The squares are summed and divided by the number of standard pulse records in the rise time sample set, and then the square root is taken to obtain the rise time standard deviation. Similarly, the fall time standard deviation and pulse width standard deviation are calculated. The arithmetic mean of the rise time standard deviation, fall time standard deviation, and pulse width standard deviation is taken to obtain the standard deviation of all standard pulse characteristic parameters in the typical partial discharge pulse database. This standard deviation is multiplied by 0.5 to obtain the preset residual threshold. Using pulse characteristic parameters from the same abnormal operating period, their actual distribution characteristics are statistically analyzed. The discrimination threshold is redefined based on the distribution tail characteristics, and a fixed multiple of 0.5 of the standard deviation is used as the residual threshold. The basis for setting this coefficient of 0.5 is that, under the condition that the pulse characteristic parameters follow a normal distribution, the 0.5 times standard deviation interval covers approximately 38% of the distribution range, which can distinguish characteristic vectors that deviate significantly from the center of the standard pulse distribution. The matching residual of each pulse in the partial discharge pulse phase distribution data is traversed. When the matching residual of a pulse is less than the preset residual threshold, the pulse is retained in the partial discharge pulse phase distribution data; when the matching residual of a pulse is greater than or equal to the preset residual threshold, the pulse is marked as a noise pulse and removed from the partial discharge pulse phase distribution data.

[0053] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for online monitoring of cable branch boxes, characterized in that, The method includes: The system collects real-time sensor data sets from multiple monitoring nodes within the cable branch box. These real-time sensor data sets include three-phase current waveform sequences, zero-sequence voltage fluctuation sequences, and partial discharge pulse phase distribution data. Based on the phase correlation characteristics between the three-phase current waveform sequence and the zero-sequence voltage fluctuation sequence, a joint operation topology among the multiple monitoring nodes is dynamically constructed. The joint operation topology includes the voltage phase angle difference matrix and the current amplitude attenuation index of each node. Extract the coupling features between the joint operation topology and the partial discharge pulse phase distribution data to generate a multidimensional electrical feature map of the cable branch box. The multidimensional electrical feature map includes the phase cluster center coordinates and the pulse amplitude distribution entropy. Spatiotemporal entropy change analysis is performed on the multidimensional electrical feature map to identify electrical abnormal events in the cable branch box. These electrical abnormal events include partial discharge phase interval shifts and sudden changes in current harmonic components.

2. The online monitoring method for cable branch boxes according to claim 1, characterized in that, The dynamic construction of the joint operation topology among the multiple monitoring nodes based on the phase correlation characteristics of the three-phase current waveform sequence and the zero-sequence voltage fluctuation sequence includes: The three-phase current waveform sequence is subjected to phase-by-phase zero-crossing detection processing to extract the instantaneous phase angle of each phase current and form a phase angle time sequence of each phase current. The zero-sequence voltage fluctuation sequence is processed to locate the peak value, and the time corresponding to the zero-crossing point of the zero-sequence voltage is extracted as the zero-sequence voltage phase reference. The phase angle time series of each phase current is compared with the zero-sequence voltage phase reference by performing a difference operation to calculate the phase difference between each phase current and the zero-sequence voltage, thereby generating the voltage phase angle difference matrix. Based on the voltage phase angle difference matrix, the attenuation coefficient of the three-phase current amplitude of each monitoring node is fitted to generate the current amplitude attenuation index.

3. The online monitoring method for cable branch boxes according to claim 2, characterized in that, The step of performing phase-by-phase zero-crossing detection processing on the three-phase current waveform sequence and extracting the instantaneous phase angle of each phase current includes: The current waveform of each phase in the three-phase current waveform sequence is binarized, and the sampling points with current amplitude greater than zero are marked as the first state, and the sampling points with current amplitude less than or equal to zero are marked as the second state. Traverse the binarized sequence of each phase current waveform and take the alternating boundary points of the continuous first state segment and the second state segment as candidate zero crossing points. For each candidate zero-crossing point, perform linear interpolation correction to calculate its precise zero-crossing time. Use the ratio of the angle between two adjacent precise zero-crossing times to the preset power frequency cycle angle as the unit increment of the instantaneous phase angle, and accumulate to generate the phase angle time series.

4. The online monitoring method for cable branch boxes according to claim 1, characterized in that, The step of extracting the coupling features of the joint operation topology and partial discharge pulse phase distribution data to generate a multi-dimensional electrical feature map of the cable branch box includes: Each pulse in the partial discharge pulse phase distribution data is mapped to the corresponding phase interval in the voltage phase angle difference matrix according to its occurrence time, so as to obtain the pulse count and pulse amplitude sequence in each phase interval; The pulse amplitude sequence within each phase interval is normalized, the pulse amplitude distribution entropy of each phase interval is calculated, and the distribution curve of the pulse amplitude distribution entropy as a function of phase is generated. Each element in the voltage phase angle difference matrix is ​​vector-combined with the current amplitude attenuation index of the corresponding phase to generate a multi-dimensional node feature vector for each monitoring node. Peak values ​​are extracted from the distribution curve of the pulse amplitude distribution entropy, and the phase coordinates corresponding to the peak values ​​are used as the coordinates of the phase cluster center.

5. The online monitoring method for cable branch boxes according to claim 4, characterized in that, The normalization process for the pulse amplitude sequence within each phase interval, and the calculation of the pulse amplitude distribution entropy for each phase interval, include: The pulse amplitude sequence within each phase interval is sorted, and the maximum and minimum values ​​in the pulse amplitude sequence are extracted to construct the amplitude interval; The pulse amplitude sequence within each phase interval is divided into several equal amplitude parts, and the number of pulses within each amplitude part is counted to generate an amplitude-frequency histogram. Calculate the proportion of pulses in each amplitude equal part of the amplitude-frequency histogram to the total number of pulses in that phase interval, multiply each proportion by the natural logarithm of that proportion, sum them up, and take the negative value to obtain the pulse amplitude distribution entropy.

6. The online monitoring method for cable branch boxes according to claim 1, characterized in that, The process of performing spatiotemporal entropy change analysis on the multidimensional electrical feature map to identify electrical anomalies in the cable branch box includes: The time-varying sequence of the multidimensional electrical feature map is sampled by a sliding window, with each sliding window covering a preset number of map time frames. The Euclidean distance between the phase cluster center coordinates between every two adjacent time frames is calculated to generate a center coordinate drift sequence. The temporal variation of the pulse amplitude distribution entropy within the sliding window is differentially processed to calculate the rate of change of the entropy value over time, thereby generating an entropy change rate sequence. The center coordinate drift sequence and the entropy change rate sequence are aligned and fused to generate a comprehensive anomaly index sequence, which is a weighted sum of the center coordinate drift sequence and the entropy change rate sequence. The time corresponding to the element in the comprehensive anomaly index sequence that exceeds the preset threshold is taken as the time of occurrence of the electrical anomaly event, and the phase cluster center coordinates at that time are extracted as the offset of the partial discharge phase interval, and the entropy change rate at that time is extracted as the abrupt change amplitude of the current harmonic component.

7. The online monitoring method for cable branch boxes according to claim 6, characterized in that, The step of aligning and fusing the center coordinate drift sequence with the entropy change rate sequence to generate a comprehensive anomaly index sequence includes: The center coordinate drift sequence is normalized by subtracting the mean of the sequence from each element and then dividing by the standard deviation of the sequence to obtain the normalized drift sequence. The entropy rate sequence is subjected to absolute value processing, and each element in the entropy rate sequence is replaced with its absolute value to obtain an absolute value entropy rate sequence. The absolute value entropy rate sequence is normalized by subtracting the mean of the sequence from each element in the absolute value entropy rate sequence and then dividing by the standard deviation of the sequence to obtain the normalized entropy rate sequence. The normalized drift sequence is added to the elements at the corresponding time points in the normalized entropy change rate sequence to generate the comprehensive anomaly index sequence.

8. The online monitoring method for cable branch boxes according to claim 1, characterized in that, The method further includes noise filtering of the partial discharge pulse phase distribution data: For each pulse in the partial discharge pulse phase distribution data, time-domain waveform features are extracted, and three time-domain feature parameters, namely pulse rise time, pulse fall time, and pulse width, are calculated. The pulse rise time, pulse fall time, and pulse width are compared one by one with the standard pulse parameters in the preset partial discharge typical pulse database, and the Euclidean distance between the characteristic parameters of each pulse and each standard pulse parameter is calculated. The minimum Euclidean distance corresponding to each pulse is used as the matching residual of that pulse. Pulses with matching residuals less than a preset residual threshold are retained, while pulses with matching residuals greater than or equal to the preset residual threshold are marked as noise pulses and removed from the partial discharge pulse phase distribution data.

9. The online monitoring method for cable branch boxes according to claim 8, characterized in that, The preset residual threshold is 0.5 times the standard deviation of all standard pulse characteristic parameters in the typical partial discharge pulse database.

10. An online monitoring system for cable branch boxes, characterized in that, The device includes a processor and a memory, the memory and the processor being connected. The memory is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the memory to implement the online monitoring method for cable branch boxes as described in any one of claims 1-9.

Images