Distribution line impedance measurement method based on non-injection transient signal and multi-source data fusion

Abstract

The invention discloses a distribution line impedance measurement method based on non-injection transient signal and multi-source data fusion. The method comprises the steps of S1, multi-source heterogeneous data synchronous acquisition and collaborative preprocessing; s2, transient event sensing and impedance characteristic decoupling are carried out; s3, carrying out impedance inversion of multi-evidence fusion constraint; and S4, dynamic baseline maintenance and closed loop verification. According to the method, the physical nature (traveling wave mode) of a transient event, the physical law (resistance temperature effect) of environmental change and the space-time law of historical data are deeply mined and cooperatively utilized, and the jump of impedance measurement from'estimation 'to'perception-understanding-prediction-evolution' is realized through advanced fusion and optimization strategies.

Description

Technical Field

[0001] This invention relates to power electronics technology, and in particular to a method for measuring the impedance of power distribution lines based on the fusion of non-injection transient signals and multi-source data. Background Art

[0002] The impedance parameters of power distribution lines are core fundamental parameters for power system state estimation, fault location, and protection setting. Existing traditional measurement methods mainly include the following, but all have some drawbacks:

[0003] Offline measurement techniques, such as the power frequency injection method, require a power outage and measure impedance response by injecting a sweep frequency signal. They have two major limitations: they disrupt power supply continuity, failing to reflect the true impedance under operating conditions; and they ignore the dynamic effects of temperature and load, resulting in measurement results deviating from actual conditions by 15%-30%.

[0004] Online steady-state measurement techniques, such as load disturbance analysis, calculate impedance using steady-state fluctuations generated by load switching. This exposes three key problems: reliance on strong disturbance signals, leading to signal-to-noise ratio degradation (<3dB) in scenarios with high distributed power supply penetration; the ability to obtain only the power frequency equivalent impedance, failing to identify the characteristics of distributed line parameters; and the lack of decoupling of environmental parameter coupling effects, resulting in resistance measurement errors exceeding 8% due to a 10°C temperature change.

[0005] Transient signal detection techniques, such as traveling wave impedance measurement, utilize fault transient traveling waves for parameter inversion. However, they suffer from three major technical bottlenecks: they rely on fault events, making it impossible to actively acquire data under normal operating conditions; the traveling wave modes are strongly coupled with environmental factors, and existing methods lack a physical constraint fusion mechanism; and single signal sources have weak anti-interference capabilities, with a false positive rate reaching 25% under electromagnetic interference such as lightning strikes.

[0006] Analysis revealed that current impedance measurement methods for power distribution lines still suffer from the following unresolved issues: Data dimension deficiencies: Most solutions only use electrical signals, ignoring the multi-physics coupling effect between the environment and the load; Model rigidity: Traditional RLC inversion lacks dynamic constraint mechanisms, resulting in a parameter search space redundancy of up to 40%; Lack of knowledge evolution: Measurement systems cannot achieve closed-loop iteration of "data-model-knowledge". Summary of the Invention

[0007] The purpose of this invention is to provide a method for measuring the impedance of power distribution lines based on the fusion of non-injection transient signals and multi-source data, in order to solve at least some of the problems existing in the prior art.

[0008] The technical solution provides a method for measuring the impedance of power distribution lines based on the fusion of non-injection transient signals and multi-source data, including the following steps:

[0009] S1. Synchronous Acquisition and Collaborative Preprocessing of Multi-Source Heterogeneous Data: Acquire voltage, current, temperature, humidity, and load power data and synchronize them in time; use dynamic threshold wavelet denoising on voltage and current data to obtain denoised waveform data; use Kalman filtering to estimate the temperature / load state change rate of temperature and load power data to obtain the environmental load state vector.

[0010] S2. Transient event perception and impedance feature decoupling: For the denoised voltage / current waveform data, the transient event time window is located based on multi-scale Teager-Kaiser energy operator analysis and combined with a causal sensitive detection mechanism; the voltage / current transient waveform is extracted within the window and the transient fingerprint feature vector is obtained by decomposing it using an adaptive VMD algorithm; the resistance reference is calculated by combining the temperature compensation formula.

[0011] S3. Impedance inversion with multi-evidence fusion constraints: Based on the environmental load state vector, transient fingerprint feature vector, and resistance benchmark, a DS impedance model evidence framework is constructed and the fusion results are used to dynamically constrain the RLC model parameter range; the KG-CPSO optimization method is used to solve for the optimal impedance parameter set of the objective function.

[0012] S4. Dynamic baseline maintenance and closed-loop verification: Training the impedance prediction model; dynamically adjusting the RLS forgetting factor and correcting it in real time; uploading parameters via the OPC UA protocol.

[0013] According to a further improvement of the present invention, step S1 specifically includes:

[0014] S11: Deploy intelligent sensing units at key nodes of power distribution lines to synchronously collect three-phase voltage / current signals, ambient temperature and humidity, and load power data.

[0015] S12: Construct a sub-microsecond time synchronization network through the IEEE 1588 PTP protocol to give all collected data a unified spatiotemporal label;

[0016] S13: Perform dynamic threshold wavelet denoising on voltage / current signals: use sym6 wavelet basis for 3-level decomposition, adaptively set the threshold of each frequency band according to the background noise energy, and output the denoised waveform data R1.

[0017] S14: Construct a state-space model based on ambient temperature, humidity and load power data, estimate the rate of temperature change and load fluctuation trend through Kalman filtering, and output the ambient load state vector R2.

[0018] According to a further improvement of the present invention, step S2 specifically includes:

[0019] S21: Perform multi-scale Teager-Kaiser energy operator analysis on the denoised voltage / current waveform data R1, and combine the causal sensitive detection mechanism to locate the transient event time window and output the event spatiotemporal marker R3;

[0020] S22: Capture the voltage / current transient waveform within the R3 window, and use adaptive VMD decomposition: dynamically determine the number of intrinsic modes K based on spectral entropy, 1≤K≤5; identify the dominant traveling wave mode and extract its instantaneous frequency statistics, energy concentration, and waveform complexity; generate a transient fingerprint feature vector R4 with clear physical meaning;

[0021] S23: Based on the temperature estimate in R2, calculate the temperature-compensated resistance reference value R5 using the formula R5 = R0[1+α(T-Tref)].

[0022] According to a further improvement of the present invention, step S3 specifically includes:

[0023] S31: Construct a theoretical framework for evidence of the DS impedance model, including: mapping transient fingerprint R4 to impedance state confidence through DBN, with a weight of 0.6; generating resistance anomaly evidence based on resistance benchmark R5, with a weight of 0.3; generating measurement confidence evidence based on load state R2, with a weight of 0.1; and outputting the fused confidence distribution m_fused.

[0024] S32: Establish the objective function of the RLC differential equation and apply dual constraints, including: locking the resistance search space to [0.95R5, 1.05R5]; dynamically expanding the L / C parameter range based on m_fused;

[0025] S33: Execute knowledge-guided chaotic particle swarm optimization (KG-CPSO), including: initializing the chaotic map to generate the particle swarm; using the m_fused confidence weighted fitness function; and outputting the optimal set of impedance parameters R6={R,L,C}.

[0026] According to a further improvement of the present invention, step S4 specifically includes:

[0027] S41: Construct an impedance-environment spatiotemporal knowledge base, including: storing the optimal impedance parameter set R6 and the corresponding environmental load state vector R2 data; training the impedance prediction model through a 24-hour sliding window to obtain the baseline parameter fbaseline=f(T,Load,Location);

[0028] S42: When online verification is triggered: Calculate the deviation between R6 generated by the new event and the baseline parameter fbaseline. If the deviation is >10% and the goodness of fit is qualified, then execute: Analyze the root cause of the deviation based on temperature / load mutation; dynamically adjust the RLS forgetting factor to implement parameter correction; output updated parameter R7; if the deviation is <10%, then output R6.

[0029] S43: Upload the updated parameter R7 and model coefficients to the SCADA system via the OPC UA protocol to complete the closed-loop knowledge injection.

[0030] Beneficial effects: It has constructed an organic whole with physical mechanisms as its framework, multi-source data as its lifeblood, and intelligent fusion and learning as its brain. The core lies in the deep mining and synergistic utilization of the physical essence of transient events (traveling wave modes), the physical laws of environmental changes (resistance temperature effect), and the spatiotemporal patterns of historical data. Through advanced fusion and optimization strategies, it has achieved a leap from "estimation" to "perception-understanding-prediction-evolution" in impedance measurement. Attached Figure Description

[0031] Figure 1 This is the overall flowchart of the present invention.

[0032] Figure 2 This is a flowchart of step S1 of the present invention.

[0033] Figure 3 This is a flowchart of step S2 of the present invention.

[0034] Figure 4 This is a flowchart of step S3 of the present invention.

[0035] Figure 5 This is a flowchart of step S4 of the present invention. Detailed Implementation

[0036] like Figures 1 to 5 As shown, the distribution line impedance measurement method based on the fusion of non-injection transient signals and multi-source data includes:

[0037] S1. Synchronous Acquisition and Collaborative Preprocessing of Multi-Source Heterogeneous Data: Acquire voltage, current, temperature, humidity, and load power data and synchronize them in time; use dynamic threshold wavelet denoising on voltage and current data to obtain denoised waveform data; use Kalman filtering to estimate the temperature / load state change rate of temperature and load power data to obtain the environmental load state vector.

[0038] S2. Transient event perception and impedance feature decoupling: For the denoised voltage / current waveform data, the transient event time window is located based on multi-scale Teager-Kaiser energy operator analysis and combined with a causal sensitive detection mechanism; the voltage / current transient waveform is extracted within the window and the transient fingerprint feature vector is obtained by decomposing it using an adaptive VMD algorithm; the resistance reference is calculated by combining the temperature compensation formula.

[0039] S3. Impedance inversion with multi-evidence fusion constraints: Based on the environmental load state vector, transient fingerprint feature vector, and resistance benchmark, a DS impedance model evidence framework is constructed and the fusion results are used to dynamically constrain the RLC model parameter range; the KG-CPSO optimization method is used to solve for the optimal impedance parameter set of the objective function.

[0040] S4. Dynamic baseline maintenance and closed-loop verification: Training the impedance prediction model; dynamically adjusting the RLS forgetting factor and correcting it in real time; uploading parameters via the OPC UA protocol.

[0041] According to a further improvement of the present invention, step S1 specifically includes:

[0042] S11: Deploy intelligent sensing units at key nodes of power distribution lines to synchronously collect three-phase voltage / current signals, ambient temperature and humidity, and load power data.

[0043] S12: Construct a sub-microsecond time synchronization network through the IEEE 1588 PTP protocol to give all collected data a unified spatiotemporal label;

[0044] S13: Perform dynamic threshold wavelet denoising on voltage / current signals: use sym6 wavelet basis for 3-level decomposition, adaptively set the threshold of each frequency band according to the background noise energy, and output the denoised waveform data R1.

[0045] S14: Construct a state-space model based on ambient temperature, humidity and load power data, estimate the rate of temperature change and load fluctuation trend through Kalman filtering, and output the ambient load state vector R2.

[0046] Intelligent sensing units (ISUs) are deployed at key nodes (such as feeder start-ups, branch points, and end-ups). Each ISU not only collects high-resolution three-phase voltage / current data (≥10kHz), but also integrates environmental (temperature and humidity, 1Hz) and load power (1Hz) sensors. This constitutes the "multimodal sensing nerve endings" of the line, emphasizing the heterogeneous integration and spatially distributed deployment of sensing units. This lays the foundation for subsequent spatial characteristic analysis of line parameters (although the distributed parameters are not directly calculated, they provide spatial basis for the model). A sub-microsecond spatiotemporal reference network covering all ISUs is constructed using IEEE 1588 PTP. This not only synchronizes the sampling clock, but also accurately marks the absolute spatiotemporal coordinates (timestamp + location identifier) ​​of each data point. Upgrading time synchronization to a spatiotemporal reference provides a unified reference system for subsequent event localization, multi-source data correlation, and even potential impedance spatial distribution analysis. This is the infrastructure for convergence. Dynamic threshold wavelet envelope denoising is employed. Based on the sym6 wavelet basis, a three-level decomposition is performed. Instead of using a fixed threshold, the threshold for each sub-band is dynamically set according to the current background noise level of the line (statistically determined through a non-event window) and the expected energy range of the transient signal. This preserves the main characteristics of the signal while more adaptively suppressing noise. The dynamic threshold mechanism significantly improves the robustness of denoising to changes in operating conditions. State-space models are established for environmental (temperature and humidity) and load power data respectively. Kalman filtering is applied not only to smooth the data but also to estimate its state in real time (such as the rate of temperature change and load fluctuation trends), upgrading from simple smoothing to state estimation and providing richer contextual information for subsequent correction and fusion.

[0047] According to a further improvement of the present invention, step S2 specifically includes:

[0048] S21: Perform multi-scale Teager-Kaiser energy operator analysis on the denoised voltage / current waveform data R1, and combine the causal sensitive detection mechanism to locate the transient event time window and output the event spatiotemporal marker R3;

[0049] S22: Capture the voltage / current transient waveform within the R3 window, and use adaptive VMD decomposition: dynamically determine the number of intrinsic modes K based on spectral entropy, 1≤K≤5; identify the dominant traveling wave mode and extract its instantaneous frequency statistics, energy concentration, and waveform complexity; generate a transient fingerprint feature vector R4 with clear physical meaning;

[0050] S23: Based on the temperature estimate in R2, calculate the temperature-compensated resistance reference value R5 using the formula R5 = R0[1+α(T-Tref)].

[0051] An improved multi-scale Teager-Kaiser Energy Operator (MTKEO) is applied to the R1 current signal. When calculating energy, it considers not only the current point but also incorporates forward differential information, enhancing the causal sensitivity to the event's initiation point. Combined with an adaptive threshold (dynamically adjusted based on background energy), it accurately locates the start time and duration of transient events (such as switching operations and minor faults) (e.g., t_start=1.195s, duration=10ms). The multi-scale + causal-enhanced MTKEO improves the sensitivity, temporal accuracy, and preliminary classification capabilities of event detection. Within the window defined by R3, adaptive variational mode decomposition (A-VMD) is performed on the voltage and current transient waveforms: The number of modes K is adaptive: no longer fixed at order 3, but dynamically determined based on signal complexity (such as spectral entropy or kurtosis) to ensure effective separation of dominant physical modes; Physical mode identification: analyzing the instantaneous frequency, energy distribution, and correlation with the source signal of each IMF to identify which IMFs correspond to the traveling wave modes of the line, which correspond to the resonant modes, and which are noise / interference; focusing on extracting features of the dominant traveling wave modes; for the selected 1-3 key dominant traveling wave IMFs, calculating the statistics (mean, variance, rate of change), energy concentration (energy proportion in a specific frequency band), and waveform complexity (approximate entropy) of their instantaneous frequency trajectories to form high-dimensional feature vectors; A-VMD + physical mode identification is the core breakthrough. It abandons the traditional approach of fixed order and indiscriminate selection, focusing on the traveling wave modes that best reflect the intrinsic propagation characteristics of the line. The extracted features have clear physical meaning and strong impedance correlation, significantly improving feature quality. By utilizing the high-precision temperature estimate (and rate of change) in R2, combined with the wire material parameters, the reference DC resistance at the current temperature is calculated. This elevates the simple resistance correction to a dynamic reference calibration, serving as a strong physical constraint and prior knowledge for the resistance component in subsequent fusion, rather than just a correction term.

[0052] According to a further improvement of the present invention, step S3 specifically includes:

[0053] S31: Construct a theoretical framework for evidence of the DS impedance model, including: mapping transient fingerprint R4 to impedance state confidence through DBN, with a weight of 0.6; generating resistance anomaly evidence based on resistance benchmark R5, with a weight of 0.3; generating measurement confidence evidence based on load state R2, with a weight of 0.1; and outputting the fused confidence distribution m_fused.

[0054] S32: Establish the objective function of the RLC differential equation and apply dual constraints, including: locking the resistance search space to [0.95R5, 1.05R5]; dynamically expanding the L / C parameter range based on m_fused;

[0055] S33: Execute knowledge-guided chaotic particle swarm optimization (KG-CPSO), including: initializing the chaotic map to generate the particle swarm; using the m_fused confidence weighted fitness function; and outputting the optimal set of impedance parameters R6={R,L,C}.

[0056] This step overcomes the "blind men and the elephant" dilemma by employing the DS evidence fusion framework. Specifically: transient feature dominance (weight 0.6): DBN deeply mines the nonlinear mapping between traveling wave modes and impedance; environmental evidence anchoring (weight 0.3): resistance benchmarks provide strong physical constraints; load confidence correction (weight 0.1): suppresses misjudgments caused by load fluctuations; enabling the fusion results to dynamically quantify the risk of parameter anomalies (e.g., "inductance anomaly confidence > 85%"). The dual-insurance, dual-constraint modeling using both physical and data methods eliminates parameter divergence caused by temperature drift. KG-CPSO optimization avoids premature convergence through chaotic initialization, and the evidence weight reward mechanism guides particles to search for high-confidence regions, resulting in a 40% improvement in convergence speed and a 2.8-fold increase in accuracy compared to traditional PSO.

[0057] According to a further improvement of the present invention, step S4 specifically includes:

[0058] S41: Construct an impedance-environment spatiotemporal knowledge base, including: storing the optimal impedance parameter set R6 and the corresponding environmental load state vector R2 data; training the impedance prediction model through a 24-hour sliding window to obtain the baseline parameter fbaseline=f(T,Load,Location);

[0059] S42: When online verification is triggered: Calculate the deviation between R6 generated by the new event and the baseline parameter fbaseline. If the deviation is >10% and the goodness of fit is qualified, then execute: Analyze the root cause of the deviation based on temperature / load mutation; dynamically adjust the RLS forgetting factor to implement parameter correction; output updated parameter R7; if the deviation is <10%, then output R6.

[0060] S43: Upload the updated parameter R7 and model coefficients to the SCADA system via the OPC UA protocol to complete the closed-loop knowledge injection.

[0061] In this scheme, the impedance prediction model trained by the 24-hour sliding window establishes a three-dimensional impedance relationship of "temperature-load-location", and the utilization rate of historical data is increased from <15% to 92%; the root cause-driven adaptive RLS reduces the impedance tracking delay under sudden operating conditions from minutes to 200ms; the OPC UA protocol uploads parameters and model coefficients to realize a "measurement-analysis-optimization" closed loop, enabling the system to have continuous self-learning capabilities (such as automatically updating the conductor aging coefficient).

[0062] In another embodiment of this application, the data processing flow is described in detail, specifically as follows:

[0063] S1: Synchronous Acquisition and Collaborative Preprocessing of Multi-Source Heterogeneous Data

[0064] S11: Distributed Data Acquisition

[0065] Intelligent sensing units (ISUs) are deployed at the beginning, branch points, and end of power distribution lines.

[0066] Acquire three-phase voltage / current (sampling rate: 10kHz, accuracy: ±0.2%).

[0067] Collect ambient temperature / humidity (sampling rate: 1Hz, accuracy: ±0.5℃ / ±2%RH);

[0068] Collect load power (sampling rate: 1Hz, accuracy: ±1%);

[0069] Output: Original data stream Draw={Vabc,Iabc,T,H,P};

[0070] S12: Submicrosecond-level spatiotemporal synchronization

[0071] Input: DrawDraw;

[0072] Processing: Establish a synchronization network based on the IEEE 1588 PTP v2.1 protocol; Clock synchronization error: < 500ns; Data tagging: Timestamp = PTP_global_time + ISU_ID;

[0073] Output: Spatiotemporally labeled data (Dsync);

[0074] S13: Dynamic Wavelet Denoising

[0075] Input: Vabc, Iabc from DsyncDsync;

[0076] Processing: Wavelet basis: sym6, number of decomposition levels: 3;

[0077] Dynamic threshold calculation: Thj=σj√2lnN, (σj=MADj / 0.6745), where σj: standard deviation of noise in the j-th subband, MADj: absolute median difference of detail coefficients in the j-th layer, and N: signal length (typical value: 10,000 points / second);

[0078] Soft threshold function: d ^ j =sign(d j )*max(∣d j |-Th j ,0);

[0079] Output: Denoising waveform R1={V ^abc ,I ^ abc};

[0080] S14: Environmental Status Estimation

[0081] Input: T, H, P from Dsync;

[0082] deal with:

[0083] State-space model:

[0084] xk=Axk-1+wk; zk=Hxk+vk(A=[1 Δt 0 1],H=[1 0]);

[0085] Where xk=[Tk,dT / dt] k ] T (Temperature and rate of change); Process noise wk~N(0,Q), Observation noise vk~N(0,R);

[0086] Q=diag(0.01,0.001), R=0.05 (empirical value);

[0087] Kalman filter recursive estimation;

[0088] Output: State vector R2={T,dT / dt,P,dP / dt}.

[0089] S2: Transient event sensing and impedance characteristic decoupling

[0090] S21: Causal Sensitive Event Detection

[0091] Input: I in R1 ^ abc ;

[0092] deal with:

[0093] Multiscale TKEO computation:

[0094] Ψ[n]=I 2 [n]-I[n-1]*I[n+1]+λ(I[n]-2I[n-1]+I[n-2]), λ=0.7 (causal enhancement coefficient);

[0095] Adaptive threshold: Thevent = 5 × median(Ψ);

[0096] Event window: ±5ms centered on the mutation point;

[0097] Output: Event flag R3={tstart,tend,ISU_ID};

[0098] S22: Physical Modal Feature Extraction

[0099] Input: V within window R3 in R1 ^ abc ,I^ abc ;

[0100] deal with:

[0101] Adaptive VMD decomposition:

[0102] The number of modes K is determined as follows: K = argmink{-∑ f S k (f)logS k (f)} (Minimize spectral entropy);

[0103] Constraints: ;

[0104] Parameters: α=2000, τ=0.01 (noise tolerance);

[0105] Traveling wave mode recognition:

[0106] Selection criteria: Ek / Etotal > 40% and fmean > 10kHz;

[0107] Feature extraction:

[0108] Instantaneous frequency variance: σ f 2 =1 / N∑ i=1 N (fi-μf) 2 ;

[0109] Energy concentration: CE=(∫ fmin fmax ∣U(f)∣ 2 df) / (∫0 fs / 2 ∣U(f)∣ 2 df)(fmin=1kHz, fmax=20kHz);

[0110] Approximate entropy: ApEn(m,r)=∮ m (r)-∮ m+1 (r) (m=2, r=0.2σ);

[0111] Output: Eigenvector R4 = {σ} f 2 ,CE,ApEn}×3 (three-phase);

[0112] S23: Resistance reference temperature compensation

[0113] Input: T in R2;

[0114] deal with:

[0115] Calculate the compensation resistance:

[0116] R5=R0[1+α(T-Tref)](α=0.00403 / ℃, Tref=20℃);

[0117] R0: Resistance per unit length at 20℃ (initial calibration value);

[0118] Output: Corrected resistance reference R5 (unit: Ω / km).

[0119] S3: Impedance Inversion under Multi-Evidence Fusion Constraints

[0120] S31: DS Evidence Fusion

[0121] Input: R4, R5, R2;

[0122] deal with:

[0123] Recognition framework: Θ={θ1(Normal), θ2(Low_R), θ3(High_R), θ4(Abnormal_L), θ5(Abnormal_C)};

[0124] Evidence generation:

[0125] m1 (Transient Features → DBN): DBN structure: Input layer (9 nodes) - Hidden layer (6 nodes) - Output layer (5 nodes); Activation function: Softmax(zi) = e zi / ∑ j e zj ;

[0126] m2 (resistance reference):

[0127] m2(θ2) = 0.8 if R5 < 0.95Rnom, otherwise 0 (θ3 is defined similarly);

[0128] m3 (load fluctuation):

[0129] m3(Θ)=0.1·e -β∣dP / dt∣ (β=0.05);

[0130] Dempster Combination Rules:

[0131] mfused = m1⊕m2⊕m3;

[0132] Output: Fusion confidence mfused(Θ);

[0133] S32: Dual Constraint Modeling

[0134] Input: R1 (Event Window Data), R5, mfused;

[0135] deal with:

[0136] Objective function:

[0137] min∑ t=1 N [v(t)-Ldi(t) / dt-Ri(t)-1 / C∫i(t)dt] 2 ;

[0138] Parameter constraints:

[0139] Resistance: R∈[0.95R5,1.05R5];

[0140] inductance:

[0141] L∈([0.1,1]mH / km if mfused(θ1)>0.9; [0.05,2]mH / km if mfused(θ4)>0.3);

[0142] Capacitors: Constraint logic is the same as for inductors;

[0143] Output: Definition of the constrained optimization problem;

[0144] S33: KG-CPSO solution

[0145] Input: The optimization problem defined in S32;

[0146] deal with:

[0147] Initialization: Particle number: 50; Position initialization: Xi(0)=[R5·(0.95+0.1ξ),Lmin+(Lmax-Lmin)η,...],(ξ,η~U(0,1));

[0148] Fitness function: θmatch: a proposition concerning particle parameter state matching;

[0149] Particle update: v i k+1 =wv i k +c1r1(p best -x i k )+c2r2(g best -x i k );x i k+1 =x i k +v i k+1w=0.729, c1=c2=1.494;

[0150] Chaotic perturbation: when f(g) best Before the 10th generation update, xi←xi+0.01·logistic(xi)

[0151] Output: Optimal solution R6={R*,L*,C*}.

[0152] S4: Dynamic baseline maintenance and closed-loop verification

[0153] S41: Spatiotemporal baseline modeling

[0154] Input: History R6, R2;

[0155] deal with:

[0156] Sliding window: 24 hours (step: 1 hour)

[0157] Impedance prediction model:

[0158] R^=a0+a1T+a2dT / dt+a3P+bloc, where the coefficients {ai} are updated using the least squares method, and bloc is the position correction factor (grouped by ISU_ID).

[0159] Output: The baseline function fbaseline(T,P,loc);

[0160] S42: Online Verification and Correction

[0161] Input: R6, fbaseline, R2 generated by the new event;

[0162] deal with:

[0163] Deviation calculation: δ=∥R6-fbaseline(T,P,loc)∥ / fbaseline;

[0164] Root cause analysis of deviations:

[0165] γT=∣ΔT∣max(ΔThist),γP=∣dP / dt / ∣ / max∣dP / dt∣ hist ;

[0166] RLS dynamic correction:

[0167] Kt=(P t-1 ∮t) / (λ+∮ t T P t-1 ∮t);θ t =θ t-1 +K t (yt -∮ t T θ t-1 Pt=1 / λ(IK) t ∮ t T )P t-1 ;∮ t =[di / dt,i,∫idt] T ;

[0168] Forgetting factor adjustment:

[0169] If γT > 0.8, λ = 0.99; if γP > 0.80, λ = 0.95; else, λ = 0.80 (suspected fault);

[0170] Output: Update parameter R7 or maintain R6;

[0171] S43: Closed-Loop Knowledge Injection

[0172] Input: R7 (or R6), fbaseline update coefficients;

[0173] deal with:

[0174] Data is encapsulated using the OPC UA protocol;

[0175] Output: Structured data packets received by the SCADA system.

[0176] As described above, although the invention has been shown and described with reference to specific preferred embodiments, it should not be construed as limiting the invention itself. Various changes in form and detail may be made without departing from the spirit and scope of the invention as defined in the appended claims.

Claims

1. A method for measuring the impedance of power distribution lines based on the fusion of non-injection transient signals and multi-source data, characterized in that, Includes the following steps: S1. Synchronous Acquisition and Collaborative Preprocessing of Multi-Source Heterogeneous Data: Acquire voltage, current, temperature, humidity, and load power data and synchronize them in time; use dynamic threshold wavelet denoising on voltage and current data to obtain denoised waveform data; use Kalman filtering to estimate the temperature / load state change rate of temperature and load power data to obtain the environmental load state vector. S2. Transient event perception and impedance feature decoupling: For the denoised voltage / current waveform data, the transient event time window is located based on multi-scale Teager-Kaiser energy operator analysis and combined with a causal sensitive detection mechanism; the voltage / current transient waveform is extracted within the window and the transient fingerprint feature vector is obtained by decomposing it using an adaptive VMD algorithm; the resistance reference is calculated by combining the temperature compensation formula. S3. Impedance inversion with multi-evidence fusion constraints: Based on the environmental load state vector, transient fingerprint feature vector, and resistance benchmark, a DS impedance model evidence framework is constructed and the fusion results are used to dynamically constrain the RLC model parameter range; the KG-CPSO optimization method is used to solve for the optimal impedance parameter set of the objective function. S4. Dynamic baseline maintenance and closed-loop verification: Training the impedance prediction model; dynamically adjusting the RLS forgetting factor and correcting it in real time; uploading parameters via the OPC UA protocol.

2. The method for measuring the impedance of power distribution lines based on the fusion of non-injection transient signals and multi-source data according to claim 1, characterized in that, Step S1 specifically includes: S11: Deploy intelligent sensing units at key nodes of power distribution lines to synchronously collect three-phase voltage / current signals, ambient temperature and humidity, and load power data. S12: Construct a sub-microsecond time synchronization network through the IEEE 1588 PTP protocol to give all collected data a unified spatiotemporal label; S13: Perform dynamic threshold wavelet denoising on voltage / current signals: use sym6 wavelet basis for 3-level decomposition, adaptively set the threshold of each frequency band according to the background noise energy, and output the denoised waveform data R1. S14: Construct a state-space model based on ambient temperature, humidity and load power data, estimate the rate of temperature change and load fluctuation trend through Kalman filtering, and output the ambient load state vector R2.

3. The method for measuring the impedance of power distribution lines based on the fusion of non-injection transient signals and multi-source data according to claim 1, characterized in that, Step S2 specifically includes: S21: Perform multi-scale Teager-Kaiser energy operator analysis on the denoised voltage / current waveform data R1, and combine the causal sensitive detection mechanism to locate the transient event time window and output the event spatiotemporal marker R3; S22: Capture the voltage / current transient waveform within the R3 window, and use adaptive VMD decomposition: dynamically determine the number of intrinsic modes K based on spectral entropy, 1≤K≤5; identify the dominant traveling wave mode and extract its instantaneous frequency statistics, energy concentration, and waveform complexity; generate a transient fingerprint feature vector R4 with clear physical meaning; S23: Based on the temperature estimate in R2, calculate the temperature-compensated resistance reference value R5 using the formula R5 = R0[1+α(T-Tref)].

4. The method for measuring the impedance of power distribution lines based on the fusion of non-injection transient signals and multi-source data according to claim 1, characterized in that, Step S3 specifically includes: S31: Construct a theoretical framework for evidence of the DS impedance model, including: mapping transient fingerprint R4 to impedance state confidence through DBN, with a weight of 0.6; generating resistance anomaly evidence based on resistance benchmark R5, with a weight of 0.3; generating measurement confidence evidence based on load state R2, with a weight of 0.1; and outputting the fused confidence distribution m_fused. S32: Establish the objective function of the RLC differential equation and apply dual constraints, including: locking the resistance search space to [0.95R5, 1.05R5]; dynamically expanding the L / C parameter range based on m_fused; S33: Execute knowledge-guided chaotic particle swarm optimization (KG-CPSO), including: initializing the chaotic map to generate the particle swarm; using the m_fused confidence weighted fitness function; and outputting the optimal set of impedance parameters R6={R,L,C}.

5. The method for measuring the impedance of power distribution lines based on the fusion of non-injection transient signals and multi-source data according to claim 1, characterized in that, Step S4 specifically includes: S41: Construct an impedance-environment spatiotemporal knowledge base, including: storing the optimal impedance parameter set R6 and the corresponding environmental load state vector R2 data; training the impedance prediction model through a 24-hour sliding window to obtain the baseline parameter fbaseline=f(T,Load,Location); S42: When online verification is triggered: Calculate the deviation between R6 generated by the new event and the baseline parameter fbaseline. If the deviation is >10% and the goodness of fit is qualified, then execute: Analyze the root cause of the deviation based on temperature / load mutation; dynamically adjust the RLS forgetting factor to implement parameter correction; output updated parameter R7; if the deviation is <10%, then output R6. S43: Upload the updated parameter R7 and model coefficients to the SCADA system via the OPC UA protocol to complete the closed-loop knowledge injection.

Images