A method and device for identifying a household line association relationship online

Abstract

This invention belongs to the field of low-voltage distribution network customer line association identification technology, and discloses an online identification method and device for customer line association relationships. The method includes: acquiring voltage and current waveforms of the customer's main inlet and each branch line using a synchronous clock; preprocessing the waveform data to filter out interference; extracting steady-state harmonic features and transient impulse features to form a multi-dimensional feature vector; calculating waveform similarity using a dynamic time warping algorithm, and combining it with a support vector machine for load identification and association matching; and automatically generating a visualized topology map containing the entire chain of association relationships between lines, switches, loads, and power supply access points. This invention achieves accurate customer line association identification under uninterrupted power supply conditions, and has advantages such as strong anti-interference capability, convenient access, and visualized results. It can effectively support emergency power supply and power safety inspections for important users such as hospitals and large commercial complexes.

Description

Technical Field

[0001] This invention belongs to the field of low-voltage distribution network customer line association identification technology, and in particular, it is a method and device for online fingerprint identification of customer line association based on load characteristics. Background Technology

[0002] The number of large users, such as hospitals and large commercial complexes, continues to increase. These users have numerous outgoing lines, each corresponding to target loads of different importance levels. In practical applications, due to reasons such as lost drawings, personnel changes, load switching, and damaged markings, the correspondence between the busbar outgoing switches and the terminal loads in the distribution room often becomes confused. In extreme power outage scenarios such as severe weather and power grid failures, the security loads of important users need to be quickly connected to backup power equipment. However, the existing method of confirming the connection between user lines mainly relies on power outage verification, which is completely unsuitable for important loads such as hospital ICUs and operating rooms where power supply cannot be interrupted.

[0003] Currently, technical solutions for confirming the relationship between customer lines are mainly divided into three categories: traditional power outage line verification technology, line-transformer relationship testing technology, and existing online identification technology. However, all of them have limitations and cannot meet the actual needs of large and important users. Traditional power outage line verification technology is currently the most widely used basic solution. It involves manually disconnecting a branch switch and observing whether the terminal load is de-energized, thereby confirming the correspondence between the switch and the load. Although it can achieve correlation confirmation in simple scenarios, its core defects are extremely prominent: First, the applicable scenarios are strict, relying entirely on power outage operations; second, the efficiency is extremely low. For large users with more than 10 branches, a single line verification requires power outage testing of each branch, which usually takes more than 4 hours and requires multiple people to operate together, resulting in high labor costs; third, the safety is poor, with the risk of arc discharge caused by misoperation during power outage and restoration, especially in harsh environments such as humid and dusty distribution rooms. Line-transformer relationship testing instruments are only suitable for simple scenarios with three or fewer branches. They are complex to operate and have poor application effects, and cannot meet the needs of large users for confirming the routing of multiple complex lines. Some online identification technologies rely on the transmission of single electrical parameters, such as voltage and current, which are susceptible to interference from power grid harmonics and load fluctuations, resulting in insufficient signal stability and a high rate of misjudgment. Furthermore, existing identification equipment suffers from poor compatibility of its access modules, failing to quickly adapt to various interfaces such as branch switches, sockets, and low-voltage switches, leading to low access efficiency and unstable connections, further impacting the smoothness of the identification process. In addition, existing technologies lack efficient means of visualizing the relationships between customer lines, meaning that identification results cannot directly guide on-site power preservation operations and are insufficient to meet the rapid response needs in emergency situations.

[0004] In summary, current technical solutions generally suffer from common defects such as difficulty in uninterrupted power-on identification, weak anti-interference capabilities, low access efficiency, and insufficient visualization of results. They cannot meet the needs of important users such as hospitals and large commercial complexes for fast, accurate, and secure identification of customer line relationships. There is an urgent need for an online identification technology that can achieve uninterrupted power-on data collection, strong anti-interference capabilities, multi-interface adaptation, and topology-based presentation of customer line relationships. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and propose an online identification method and device for customer line association based on load characteristic waveform recognition. This method enables accurate identification of customer line association in complex internal circuits of large users under uninterrupted power supply conditions. It also solves the problems of inconvenient access, poor anti-interference ability of identification, and insufficient visualization of results, thus providing efficient technical support for emergency power supply.

[0006] The technical solution adopted by this invention to solve the technical problem is:

[0007] The first aspect of the present invention is to provide an online identification method for user line association relationships, comprising the following steps:

[0008] The system collects voltage and current waveform data from the main user access point and each branch line, and achieves high-precision synchronization between the acquisition units through a synchronous clock.

[0009] The acquired waveform data is preprocessed to filter out power grid harmonics and noise interference;

[0010] Extract the steady-state and transient features of the preprocessed waveform data to form a multidimensional feature vector of the load;

[0011] Based on the multidimensional feature vector, the correlation between each branch line and the terminal load is identified by waveform similarity calculation;

[0012] Based on the identification results, a user line association topology diagram is automatically generated and output.

[0013] Furthermore, the extraction of the steady-state features includes: performing a windowed Fourier transform on the current waveform of multiple consecutive power frequency cycles, extracting the amplitude and phase parameters of the fundamental wave and each harmonic, and calculating the total harmonic distortion rate.

[0014] Furthermore, the extraction of transient features includes: monitoring the rate of change of current to identify load switching events, capturing the original waveforms before and after the events, and extracting their time-frequency domain features using wavelet transform.

[0015] Furthermore, the waveform similarity calculation employs a dynamic time warping algorithm, combined with a support vector machine classifier for feature matching and association determination.

[0016] Furthermore, the identification of the correlation between each branch line and the terminal load includes: firstly, performing preliminary screening based on steady-state feature vectors to narrow down the range of candidate branches; and then performing final matching and confirmation of the screened candidate branches based on transient feature vectors.

[0017] The second aspect of this invention is to provide an online identification device for customer line relationships using the aforementioned method. The modular relationship of this device is controlled by a portable host computer. Each functional module collaborates through a logical chain of signal acquisition, data processing, intelligent identification, and result presentation. The overall workflow covers the entire process of deployment, acquisition, analysis, and output. The device includes a portable host computer, a waveform acquisition module, a fast access module, a data processing module, a waveform recognition module, and a topology drawing module. These modules work together to achieve online identification and visualization of customer line relationships.

[0018] The portable host unit serves as the control core, responsible for data processing and result display.

[0019] Multiple waveform acquisition modules are set up at the main user input and each branch line to synchronously acquire voltage and current waveform data;

[0020] The quick-connect module is used to quickly and reliably connect the waveform acquisition module to different types of electrical interfaces;

[0021] The data processing module is used to filter and extract features from the acquired waveform data;

[0022] The waveform recognition module is used for load identification and correlation matching based on feature vectors.

[0023] The topology drawing module is used to automatically generate a user line association topology diagram based on the recognition results.

[0024] Furthermore, the quick access module adopts a snap-fit ​​housing structure with built-in elastic probes to adapt to one or more interface types among sockets, circuit breaker terminals, and branch switch terminals.

[0025] Furthermore, the waveform acquisition module includes an open-type Rogowski coil, a high-precision voltage transformer, and an AD chip with a synchronous sampling clock, supporting multi-channel synchronous sampling.

[0026] Furthermore, the data processing module uses the RLS adaptive filtering algorithm to filter out interference from the original waveform, and combines FFT and wavelet transform to extract steady-state and transient features.

[0027] Furthermore, the topology drawing module generates a single-run diagram in SVG format, which supports graphic scaling and panning, and marks the association between lines, switches, loads and power supply access points.

[0028] Furthermore, the portable host and each waveform acquisition module achieve data transmission and synchronous control through wireless communication.

[0029] The advantages and positive effects of this invention are:

[0030] 1. This invention uses a non-invasive waveform acquisition and analysis method, employing a wideband sensor to capture richer waveform details, without requiring power outages throughout the process. It perfectly solves the pain point of critical users such as hospital ICUs and large commercial buildings where power supply cannot be interrupted, thus avoiding losses and risks caused by power outages.

[0031] 2. This invention overcomes the limitations of identifying single electrical parameters by simultaneously extracting the transient impact characteristics and steady-state harmonic characteristics of the load, forming a multi-dimensional feature vector, which significantly improves the robustness and accuracy of the identification. The waveform characteristics of different loads differ significantly, making it less susceptible to interference from power grid harmonics and load fluctuations, and resulting in high accuracy.

[0032] 3. This invention designs an integrated quick-access module consisting of a snap-fit ​​housing, an elastic probe, and an insulating protective layer, structurally solving the pain points of inconvenient access and unstable connection of existing identification devices. The housing is made of flame-retardant ABS material, and the built-in retractable copper alloy elastic probe is compatible with three types of core interfaces: 10A / 16A two-hole / three-hole sockets, 1P / 2P circuit breaker terminals, and branch switch wiring terminals, without the need to disassemble the original wiring.

[0033] 4. This invention has a built-in topology drawing module that automatically generates a single-run diagram in SVG format based on the recognition results. This transforms abstract recognition data into intuitive visual graphics, presenting the entire chain of connections between lines, switches, loads, and power supply access points. In emergencies, it can quickly locate the switches and access points corresponding to important loads, shortening the power supply response time and solving the problem that the recognition results of existing technologies are difficult to apply directly.

[0034] 5. This invention uses a portable host as its core, which is small in size and light in weight. It does not require professional and complicated operation, reduces manpower input, and improves the efficiency and accuracy of power safety inspection and customer-side power supply protection work. Attached Figure Description

[0035] Figure 1 This is a structural block diagram of the device of the present invention;

[0036] Figure 2 This is a flowchart of the identification method of the present invention;

[0037] Figure 3 This is a standard waveform diagram for a resistive load.

[0038] Figure 4 This is a standard waveform diagram for inductive loads.

[0039] Figure 5This is a standard waveform diagram for a nonlinear load.

[0040] Figure 6 This is a standard waveform diagram for a composite load. Detailed Implementation

[0041] The present invention will be further described in detail below through specific embodiments. The following embodiments are merely descriptive and not limiting, and should not be used to limit the scope of protection of the present invention.

[0042] I. System Composition and Hardware Implementation:

[0043] Reference Figure 1 The system structure diagram of this invention, shown, illustrates an online identification device for customer line association based on load characteristic waveform recognition. This device includes a portable host, multiple waveform acquisition modules, a fast access module, a data processing module, a waveform recognition module, and a topology drawing module. These modules work collaboratively, with the portable host as the control core. The waveform acquisition module acquires field data through the fast access module, which, after filtering by the data processing module, transmits it to the waveform recognition module for association determination. Finally, the topology drawing module generates a visual topology map, forming a closed loop of physical connection → data acquisition → anti-interference processing → intelligent recognition → result presentation. This achieves the goal of uninterrupted, high-precision, and easy-to-operate customer line association identification.

[0044] (1) Portable host

[0045] As the control core, it adopts an industrial-grade tablet PC (such as a 15-inch touchscreen display, sunlight-readable, supporting multi-touch), with a built-in embedded operating system, storage capacity ≥128GB, supporting WiFi / Bluetooth wireless communication, wireless synchronization accuracy ±1μs, and data transmission rate ≥10Mbps. The host is responsible for receiving waveform data uploaded by each acquisition unit, running the core algorithm, and displaying the recognition results.

[0046] (2) Waveform acquisition module

[0047] It includes a slave unit and a signal generator unit, both with identical hardware structures, differing only in software configuration. The slave unit is installed at the branch outgoing switches in the power distribution room, while the signal generator is installed on the terminal load side (such as a socket).

[0048] Sensor configuration: Current measurement uses an open-type Rogowski coil, which is non-magnetic saturation and has a wide bandwidth (DC-10kHz), making it suitable for capturing transient impulses and harmonics; voltage measurement uses a high-precision voltage transformer to ensure accurate acquisition of voltage waveforms.

[0049] Acquisition chip: ADS1256, 24-bit resolution, voltage range 0-300V (accuracy class 0.5), current range 0-100A (accuracy class 0.5), phase error <0.1°, inter-channel synchronization delay <10ns, sampling rate adjustable from 1-100kHz. All acquisition units are connected through a single synchronization signal line to achieve high-precision synchronous sampling (synchronization error <1μs).

[0050] Power supply and battery life: Low power consumption design, 5V power supply, battery life ≥ 8 hours.

[0051] (3) Quick access module

[0052] It adopts an integrated structure of snap-fit ​​shell, flexible probe, and insulating protective layer. The shell is made of flame-retardant ABS material, and the flexible probe is made of silver-plated copper alloy to ensure conductivity and oxidation resistance. This module is compatible with three core interfaces:

[0053] 10A / 16A two-hole / three-hole socket (socket adapter version);

[0054] 1P / 2P circuit breaker terminal blocks (18mm / 36mm width, circuit breaker compatible).

[0055] Branch switch terminals (4-8mm in diameter).

[0056] It automatically snaps into place after insertion, eliminating the need to disassemble the original wiring and improving connection efficiency by over 80%.

[0057] (4) Data processing module

[0058] The core DSP chip is a TMS320F28335 (150MHz, supporting floating-point operations), which communicates with the waveform acquisition module through the SPI interface to receive raw waveform data (each frame contains sampling timestamp, current value, voltage value, hexadecimal format).

[0059] Filtering algorithm: An RLS (Recursive Least Squares) adaptive filter is adopted with a forgetting factor λ=0.99 and the covariance matrix is ​​updated recursively. It effectively filters out power grid harmonics and electromagnetic interference. After filtering, the fundamental purity is ≥95% and the harmonic distortion rate is ≤2%.

[0060] Feature extraction: The system interacts with the portable host via the EMIF interface, and the host runs FFT and wavelet transform algorithms to extract steady-state and transient features.

[0061] (5) Waveform recognition module

[0062] Built into a portable host, it uses the Dynamic Time Warping (DTW) algorithm to calculate waveform similarity and combines it with a Support Vector Machine (SVM) classifier for load type identification. The SVM uses the RBF kernel function with a penalty coefficient C=10 and gamma=0.1, and has been optimized through extensive sample training. The feature fingerprint database includes 10 typical load types, such as incandescent lamps, air conditioner compressors, ICU ventilators, and supermarket composite loads. Typical waveforms can be found [link to waveform analysis]. Figures 3-6 .

[0063] (6) Topology drawing module

[0064] Based on the recognition results, an SVG format single-run diagram is automatically generated, clearly marking the entire chain of relationships between lines, switches, loads, and power supply connection points: different branches are marked with different colors; important loads (such as ICU ventilators) are marked in red, and ordinary loads are marked in gray; power supply connection points are marked in yellow; branch switch numbers correspond one-to-one with the corresponding load names. The diagram supports scaling and panning and can be exported as PDF or PNG format.

[0065] II. Implementation Steps of Online Identification Method

[0066] Reference Figure 2 The flowchart of the identification method shown indicates that the method of the present invention includes the following steps:

[0067] Step 1: System Deployment and Synchronous Sampling

[0068] Place the portable host (master station) near the main switch in the power distribution room.

[0069] Multiple waveform acquisition modules (slave units) are installed on the cables of the branch outgoing switches that need to be identified via a quick access module (for example, 8 slave units are connected to the A-phase current terminals of 8 branch switches).

[0070] Multiple signal generator units are plugged into the power sockets of each terminal load via a quick access module (e.g., 12 signal generators for 12 loads).

[0071] The master station's synchronization function is activated, and a precise clock synchronization signal is sent to all acquisition units via wireless signal to ensure that the clock error of all units is within the microsecond range.

[0072] The sampling rate is dynamically adjusted based on the line length. When the line length is ≤1km, the sampling rate is 1kHz; when 1km < line length ≤2km, the sampling rate is 3kHz; and when the line length >2km, the sampling rate is 5kHz. The sampling rate adjustment is achieved by controlling the SPS pin of the ADS1256 via the DSP chip's I / O port, ensuring that waveform details are not lost on long lines.

[0073] Step 2: Creating Disturbance and Data Acquisition

[0074] Artificially creating load switching events: for example, suddenly turning on the target load from the off state, or turning off the running equipment, resulting in a clear step change in current.

[0075] The current sensor on the main circuit will immediately detect a change in current. Sensors on all branches will also collect data synchronously, and all units will simultaneously record the high-frequency current waveform and timestamp.

[0076] Synchronization of voltage and current waveforms is ensured by using a synchronized sampling clock. Let the current sampling sequence be... and voltage sampling sequence are sampling frequency power frequency The synchronization error must be less than 1μs, which is guaranteed by the synchronization signal line of the AD chip.

[0077] Step 3: Signal Preprocessing

[0078] The original sampled signal contains power grid harmonics, electromagnetic interference, and measurement noise. Interference removal needs to be achieved through RLS (Recursive Least Squares) adaptive filtering. The recursive covariance matrix calculation is the core of the efficient operation of the RLS algorithm. Using the RLS adaptive filter, its convergence speed is significantly better than the LMS algorithm. Current time... n filter weight vector Updated formula:

[0079]

[0080] The Kalman gain update formula is as follows:

[0081] In the formula: For the previous moment n -1 filter weight vector; Kalman gain; This represents the error signal at the current moment; Let be the covariance matrix of the previous time step; For the current moment n The input signal vector; λ Forgetting factor; T This is the transpose matrix, which converts column vectors into row vectors.

[0082] λ A value of 0.99 is chosen to achieve a balance between tracking capability and steady-state error. λ When the value approaches 1, historical data is given higher weight, and the steady-state error is smaller. λ As the value approaches zero, it focuses more on new data, resulting in stronger tracking capabilities.

[0083] Recurrence relation of covariance matrix:

[0084] In the formula: The covariance matrix at the current moment represents the fluctuation degree and linear correlation of each dimension of the sampled signal, providing a statistical basis for weight updates. By using a recursive method, the recalculation of all historical data is avoided. Updates can be made based only on the previous moment and the current sampled value, reducing the computational complexity from full calculation to recursive calculation, thus meeting the computing power requirements of real-time filtering on site.

[0085] Verify the filtering effect: The fundamental purity of the filtered current signal must be ≥95%, and the harmonic distortion rate (harmonics not generated by the load itself) must be ≤2%. Otherwise, adjust the RLS forgetting factor or the initial parameters of the covariance matrix and re-execute the preprocessing.

[0086] Step 4. Feature Extraction

[0087] All branch monitoring units upload the high-frequency current waveform data and its timestamp to the main station unit via a wireless network. The main station unit runs an algorithm to extract the feature fingerprints generated by this operation from the waveform data of the main line and all branches, and analyzes the waveforms at the moment of load start-up or shutdown. This invention establishes a standard feature fingerprint library containing 10 types of typical loads, of which the standard characteristic waveforms of 4 core load types are as follows: Figures 3-6 As shown in the figure, the current waveform characteristics of different types of loads during steady-state operation and switching are as follows: the waveform of resistive load is close to a standard sine wave with low harmonic content; the waveform of inductive load has obvious phase lag characteristics; the nonlinear load contains high-frequency spikes and high harmonic components; the composite load contains multiple characteristics such as harmonic distortion, transient impact and phase shift.

[0088] Figure 3 This is a standard waveform diagram for a resistive load (such as an incandescent lamp). A purely resistive load exhibits a standard sine wave current waveform during operation, with no significant impact distortion during transient switching, and its total harmonic distortion (THD) remains at a low level during steady-state operation, approximately 4.3%. Identification of this type of load primarily relies on its smooth sinusoidal current envelope.

[0089] Figure 4 This is a standard waveform diagram for an inductive load (such as an air conditioner compressor). During the transient phase of motor startup, the current waveform of this type of load exhibits a severe starting spike distortion at the beginning, lasting approximately 0-2 ms. After entering steady-state operation, although the current waveform returns to a sinusoidal state, the steady-state characteristics extracted by the algorithm show a significant inductive phase shift, with the current lags the voltage by approximately 45°. The strong "transient spike" and "steady-state phase shift" at the current end constitute its core identification fingerprint.

[0090] Figure 5This is a standard waveform diagram for a nonlinear load (such as an ICU ventilator). These devices often employ a rectifier-inverter structure, resulting in highly irregular pulse cluster distortion in their current waveform. In steady-state characteristics, they are rich in high-order harmonic components, with the third harmonic accounting for up to 15%. These high-frequency distortion features can be accurately extracted using windowed FFT transform.

[0091] Figure 6 This is a standard waveform diagram for a composite load. When multiple electrical devices are operating simultaneously on a branch line, the current waveform exhibits highly mixed and irregular characteristics. Randomly distributed distorted peaks are superimposed on the waveform envelope, and the overall parameters show a comprehensive phase shift, with the current lags the voltage by approximately 15°.

[0092] This invention utilizes the differences in time and frequency domains of various load waveforms to construct load feature fingerprints and achieve branch association identification. Motor-type loads exhibit a large inrush current during startup; its peak value, rise slope, and duration are analyzed. Switching power supply-type loads show a unique current spike and oscillation during startup. The waveform changes after the load stabilizes are analyzed, and the changes in the 3rd, 5th, and 7th harmonic content before and after operation are compared. Feature extraction combines FFT and wavelet transform to form a multidimensional fingerprint vector.

[0093] Steady-state feature extraction: Apply a Hanning window to the current waveform for M consecutive power frequency cycles (M is set to 10) and calculate the Discrete Fourier Transform (DFT).

[0094] Hanning window function :

[0095] In the formula: N c This represents the number of sampling points per cycle. g The sampling point number, .

[0096] Performing a DFT on the windowed current sequence yields the complex amplitude values ​​of each frequency component, whose discrete form is:

[0097]

[0098] In the formula: This represents the number of FFT transform points, with a value of 2048. k This refers to the frequency point number, which corresponds to the actual frequency. sampling frequency ;Pick k =1 corresponds to the fundamental frequency, and 2~15 correspond to the 2nd~15th harmonics; This is a current sampling sequence; j It is the imaginary unit.

[0099] Extracting the fundamental 50Hz amplitudeA 1. Fundamental 50Hz phase and the amplitude of the 2nd–15th harmonics Calculate the total harmonic distortion (THD):

[0100]

[0101]

[0102] In the formula: arg is the complex argument function; To take the modulus of a complex number; X (1) is the complex amplitude value (frequency domain) corresponding to the first frequency point.

[0103]

[0104] In the formula: X ( h ) is the first h The complex amplitude value (frequency domain) corresponding to each frequency point, .

[0105]

[0106] The steady-state eigenvector is

[0107] In the formula: These are steady-state eigenvectors; A 2- A 15 The amplitude values ​​are for the 2nd to 15th harmonics.

[0108] Transient feature extraction: Real-time monitoring of current change rate, when When a load switching transient event occurs, it is determined to be a load switching transient event. The original waveforms of two power frequency cycles before and after the event are automatically locked as transient analysis data segments. The transient waveform is captured, and eight transient feature parameters, including wavelet coefficient energy, peak current, and rise time, are extracted. A 16-dimensional feature vector is constructed, and the load switching event is captured through wavelet transform.

[0109] The transient current waveform is subjected to continuous wavelet transform (CWT). The time-domain expression of the Morlet mother wavelet is as follows:

[0110]

[0111] In the formula: t It is the time-domain independent variable; f c The center frequency of the Morlet mother wavelet. The Gaussian window width parameter ensures that the wavelet has both good temporal locality and frequency domain resolution.

[0112] Continuous wavelet transform The defining formula is:

[0113]

[0114] In the formula: a For scale parameters, For translation parameters, t It is the time-domain independent variable; It is a continuous transient current signal. The complex conjugate of the Morlet mother wavelet function; This is an energy normalization factor to ensure that the energy of wavelet coefficients is comparable at different scales.

[0115] The coefficient matrix obtained from wavelet transform Eight core transient features were extracted:

[0116] 1) Peak current: The maximum current amplitude of the transient waveform. ;2) Current from (Peak current) rises to The duration of (peak current) is expressed as the rise time. t r 3) Wavelet coefficient energy: Calculate the energy of wavelet coefficients at scales a = 0.5~4 (corresponding to frequencies 10Hz~80Hz). ,Pick and corresponding scale a max ;4) Oscillation decay time: The amplitude of the current oscillation decays to after the transient impact. Duration of (peak current) t d ;5) Five high-frequency wavelet coefficient quantiles: take the 25%, 50%, and 75% quantiles and peak-to-peak values ​​of the wavelet coefficients at scales of a=2~4 to characterize the distribution characteristics of high-frequency impacts.

[0117] Extract the wavelet coefficient energy of the waveforms before and after the event as a feature vector:

[0118]

[0119] In the formula: For feature vectors, This is the peak current. t r Rising time For the maximum wavelet coefficient energy, a max The scale parameter corresponding to the maximum energy. t d For the oscillation decay time, Quantity value P pp Peak-to-peak value, The energy of wavelet coefficients at different scales.

[0120] Step 5. Fingerprint matching and association recognition

[0121] The master station aligns the total waveform with all branch waveforms on the timeline using precise timestamps. Any observed change in the total waveform must be caused by changes in one or more branches; the goal is to find the unique source. The master station calculates the similarity between the total waveform and each branch waveform. Fingerprint recognition uses a Dynamic Time Warping (DTW) algorithm to calculate waveform similarity, ensuring optimal similarity even with slight timeline shifts. The branch with the highest similarity to the total waveform is the branch containing the target load, and the system records this association.

[0122] Let the reference fingerprint sequence be set and test sequence DTW searches for the optimal regular path to minimize the cumulative distance:

[0123]

[0124]

[0125] In the formula: The distance is Euclidean, i.e., the total path length is... p The first feature point and the branch q The Euclidean distance between each feature point is a scalar value. For the reference sequence number p One feature point, For the test sequence number q One feature point; R For reference fingerprint sequence, S For testing sequences; d The dimension index ranges from 1 to 34, representing the 34th dimension of the feature vector. d One dimension; Reference feature point In the d Feature values ​​in each dimension Reference feature point In the d The feature values ​​in each dimension are set to a similarity threshold. , D DTW If the value is below the threshold, it is considered a matching association.

[0126] Combined feature vectors:

[0127] Combining the SVM classifier, the decision function is:

[0128] In the formula: F For combined feature vectors; These are steady-state eigenvectors; These are transient feature vectors; Ns This represents the number of training samples; For Lagrange multipliers; K (,) represents the kernel function; y i For sample labels, the first i The true category of each training sample , ; b This is the bias term; sign is the sign (positive or negative) of the result within the parentheses, mapping it to +1 or -1.

[0129] RBF kernel function:

[0130] in: F i Let be the feature vector of the i-th training sample; exp be the natural exponential function; || || 2 The square of the Euclidean distance; This is a parameter for the RBF kernel function, with a value of 0.1.

[0131] Combine feature vectors Input a pre-trained SVM classifier with model parameters consistent with the technical solution. The classifier output is 1 to indicate a match to the target load type and -1 to indicate a mismatch.

[0132] Step 6. Topology Graph Generation and Application

[0133] Repeat steps 3 to 5 for all critical loads within the user's premises to establish a complete load-branch correspondence database. Based on all identified relationships and the actual switching sequence of the distribution box, the built-in software of the main station unit automatically generates the primary system wiring diagram for the user's premises, clearly marking the load connected to each switch. The generated wiring diagram can be displayed on a screen or transmitted via Bluetooth or Wi-Fi to a mobile phone, tablet, or printed out for use by power supply personnel or safety inspectors.

[0134] Different branches of the line are distinguished by different colors to avoid confusion between multiple branches; important loads are marked in red and ordinary loads are marked in gray to highlight the priority of power supply; power supply access points are marked in yellow to clearly indicate the access location of emergency power supply; the branch switch number and the corresponding load name are marked to achieve a one-to-one correspondence between switch and load.

[0135] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several modifications and improvements can be made without departing from the inventive concept, and these all fall within the protection scope of the present invention.

Claims

1. A method for online identification of user line association relationships, characterized in that, Includes the following steps: The system collects voltage and current waveform data from the main user access point and each branch line, and achieves high-precision synchronization between the acquisition units through a synchronous clock. The acquired waveform data is preprocessed to filter out power grid harmonics and noise interference; Extract the steady-state and transient features of the preprocessed waveform data to form a multidimensional feature vector of the load; Based on the multidimensional feature vector, the correlation between each branch line and the terminal load is identified by waveform similarity calculation; Based on the identification results, a user line association topology diagram is automatically generated and output.

2. The method according to claim 1, characterized in that, The extraction of steady-state features includes: performing windowed Fourier transform on the current waveform of multiple consecutive power frequency cycles, extracting the amplitude and phase parameters of the fundamental wave and each harmonic, and calculating the total harmonic distortion rate.

3. The method according to claim 1, characterized in that, The extraction of transient features includes: monitoring the rate of change of current to identify load switching events, capturing the original waveforms before and after the events, and extracting their time-frequency domain features using wavelet transform.

4. The method according to claim 1, characterized in that, The waveform similarity calculation uses a dynamic time warping algorithm, combined with a support vector machine classifier for feature matching and association determination.

5. The method according to claim 1, characterized in that, The identification of the association between each branch line and the terminal load includes: First, a preliminary screening is performed based on the steady-state feature vectors to narrow down the range of candidate branches; Then, the candidate branches after screening are finally matched and confirmed based on the transient feature vectors.

6. An online identification device for user line association relationship according to any one of claims 1 to 5, characterized in that, include: The portable host unit serves as the control core, responsible for data processing and result display. Multiple waveform acquisition modules are set at the main user input and each branch line to synchronously acquire voltage and current waveform data. The quick access module is used to quickly and reliably connect the waveform acquisition module to different types of electrical interfaces; The data processing module is used to filter and extract features from the acquired waveform data; The waveform recognition module is used for load identification and correlation matching based on feature vectors. The topology drawing module is used to automatically generate a user line association topology diagram based on the recognition results.

7. The apparatus according to claim 6, characterized in that, The quick access module adopts a snap-fit ​​housing structure and has built-in elastic probes to adapt to one or more interface types among sockets, circuit breaker terminals, and branch switch terminals.

8. The apparatus according to claim 6, characterized in that, The waveform acquisition module includes an open-type Rogowski coil, a high-precision voltage transformer, and an AD chip with a synchronous sampling clock, supporting multi-channel synchronous sampling.

9. The apparatus according to claim 6, characterized in that, The data processing module uses the RLS adaptive filtering algorithm to filter out interference from the original waveform, and combines FFT and wavelet transform to extract steady-state and transient features.

10. The apparatus according to claim 6, characterized in that, The topology drawing module generates a single-run diagram in SVG format, which supports graphic scaling and panning, and marks the relationship between lines, switches, loads and power supply access points.

Images