Fault Line Selection Method Based on Injected Current Distribution Characteristics in Distribution Networks

The distribution network, as the “last mile” connecting the power system to users, directly affects social production and life. Single-phase grounding faults are the most common type of fault in distribution networks (accounting for over 70%), but traditional fault selection methods (such as zero-sequence current comparison based on steady-state quantities and energy methods based on transient quantities) have limitations in complex scenarios (such as high-resistance grounding, arc suppression coil compensation, and parallel operation of multiple power sources). For example, during high-resistance grounding, the zero-sequence current is weak and difficult to detect, and the arc suppression coil can obscure the characteristics of the steady-state zero-sequence current.

In this context, the fault line selection method based on injected current distribution characteristics has gained widespread attention due to its unique advantages. This method actively injects specific characteristic current signals into the distribution network (or utilizes transient currents generated by the fault itself) to achieve precise line selection based on the distribution differences of current in faulted and non-faulted lines. The core logic is that the current distribution of the faulted line is significantly different from that of the non-faulted line, influenced by the fault point impedance, network topology, and parameters. By extracting this differential characteristic, line selection can be completed.

This article will elaborate on the method’s principles, key steps, technical details, and engineering applications.

2. Mechanism and Classification of Injected Current Generation

Injected current refers to specific signals artificially introduced into the distribution network to enhance fault characteristics, or transient currents generated during the fault process. Based on the generation method, it can be divided into active injected current and fault self-excited current.

2.1 Active Injected Current

Active injected current is generated by applying external excitation (such as DC, low-frequency AC, or pulse signals) to specific nodes (such as busbars or sectional switches). Common injection methods include:

• Low-frequency signal injection method: Injecting current signals at frequencies far below the power frequency (e.g., below 50Hz, typical values are 10Hz, 20Hz) at the busbar, utilizing the impedance difference between faulted and non-faulted lines at this frequency (the ground impedance of the faulted line significantly decreases due to grounding faults) to achieve current distribution separation.
• S injection method (Signal Injection): By detecting the compensation current of the arc suppression coil, injecting a signal that is out of phase with the arc suppression coil current at the neutral point to offset the compensation effect of the arc suppression coil on the steady-state zero-sequence current, thereby amplifying the fault current characteristics.
• Pulsed current injection method: Injecting a short-duration (e.g., a few milliseconds) high-amplitude pulsed current into the busbar, utilizing the propagation differences of the pulse signal in faulted and non-faulted lines (the faulted line has a low-resistance path due to grounding, causing the pulse to attenuate faster) to identify faults.

The advantages of active injected current are that the signals are controllable and the characteristics are clear, making it suitable for arc suppression coil compensation systems; however, the disadvantage is that it requires additional injection equipment and may be interfered with by operating modes (such as distributed power access).

2.2 Fault Self-Excited Current

Fault self-excited current is the transient current generated during a fault due to natural processes such as unstable arcs and energy release (e.g., high-frequency oscillating currents during arc grounding). Its characteristics include large amplitude (up to several amps to several tens of amps), high frequency (from thousands to tens of kilohertz), and distribution characteristics closely related to the fault location and transition resistance.

For example, during a single-phase grounding fault, the intermittent extinguishing and reignition of the fault point arc can excite high-frequency transient traveling waves. These traveling waves will reflect and refract due to line length and branch nodes as they propagate through the distribution network, resulting in significant differences in amplitude, polarity, and frequency components of transient currents in different lines. Fault self-excited current does not require additional injection equipment, but it is necessary to address the issues of capturing transient signals and extracting characteristics.

3. Mathematical Modeling and Analysis of Current Distribution Characteristics

To achieve the identification of current distribution characteristics, it is first necessary to establish a current distribution model of the distribution network under fault conditions.

3.1 Network Topology and Parameter Model of the Distribution Network

The distribution network can be abstracted as an undirected graph <span>G=(V,E)</span>, where <span>V</span> is the set of nodes (busbars, loads, power sources), and <span>E</span> is the set of branches (feeder segments, transformers). Each branch <span>e<span>i</span></span> can be represented as:<span>Z</span><code><span><span>i</span></span> = R<span><span>i</span></span> + jX<span><span>i</span></span> where <span>R</span><code><span><span>i</span></span> and <span>X</span><code><span><span>i</span></span> are the resistance and reactance of the branch; the set of branches connected to node <span>v<span>j</span></span> is denoted as Γ<span>(j)</span>.

When a single-phase grounding fault occurs (assumed to be phase C grounding), the fault point <span>F</span> will short-circuit the busbar <span>v<span>F</span></span> to the ground, forming a zero-sequence loop. At this time, the current distribution of each branch in the zero-sequence network is determined by the zero-sequence impedance. For non-faulted lines (lines not directly connected to the fault point), their zero-sequence current is determined by the zero-sequence voltage at the busbar and their own zero-sequence impedance; for faulted lines (lines directly connected to the fault point), their zero-sequence current is the fault point’s zero-sequence voltage divided by the sum of the fault line’s zero-sequence impedance and the system’s zero-sequence impedance.

4. Steps for Identifying Current Distribution Characteristics

The fault line selection method based on injected current distribution characteristics can be divided into four core steps: signal injection/collection → data preprocessing → feature extraction → classification decision.

4.1 Signal Injection and Collection

• Injection strategy: If active injection is used, the type and frequency of the injected signal should be selected based on the operating mode of the distribution network (e.g., whether there is an arc suppression coil). For arc suppression coil compensation systems, low-frequency signals should be selected (to avoid overlap with power frequency zero-sequence currents); non-arc suppression coil systems can use pulse signals (utilizing transient characteristics). The injection location is usually chosen at the busbar (covering all outgoing lines) to ensure the signal can reach all possible faulted lines.
• Synchronous collection: Current signals need to be synchronously collected at the heads of each outgoing line (or key nodes), with the sampling frequency meeting the Nyquist criterion (for transient signals, it should be ≥10 times the highest frequency, e.g., 50kHz). The synchronization error should be controlled within 1μs to avoid phase deviation affecting feature extraction.

4.2 Data Preprocessing

The collected raw signals contain a lot of noise (such as electromagnetic interference and load fluctuations) and need to be preprocessed using the following methods:

• Denoising: Using wavelet soft-threshold filtering (for transient signals) or moving average filtering (for steady-state signals) to remove high-frequency noise.
• Normalization: Eliminating the differences in transformation ratios of current transformers (CTs) in each branch, converting the current to per unit (based on the rated current of the busbar).
• Alignment: Aligning the starting moments of signals from each node using cross-correlation algorithms (especially for transient signals).

4.3 Feature Extraction

Feature extraction is the core of distinguishing faulted lines from non-faulted lines and needs to be designed based on the steady-state or transient characteristics of current distribution.

4.3.1 Steady-State Features (Active Injection Method)

• Amplitude difference: The injected current amplitude in the faulted line is <span> significantly greater than that in the non-faulted line.</span>
• Phase difference: The current in the faulted line is in the same phase as the injected current (during low-resistance grounding), while the current in the non-faulted line lags behind the injected current due to higher ground impedance (the specific lag angle is related to the line impedance angle).
• Energy entropy: Calculating the energy distribution entropy values of currents in each line, the faulted line has a lower entropy value due to energy concentration.

4.3.2 Transient Features (Self-Excited Current Method)

• Polarity reversal: The transient current in the faulted line will exhibit polarity reversal after reflection at the fault point (opposite to the polarity of the current at the head of the non-faulted line), which can be identified by detecting changes in the current sign at adjacent sampling points.
• High-frequency component ratio: The amplitude of the high-frequency components (e.g., above 10kHz) of the transient current in the faulted line is 2 to 5 times that of the non-faulted line, which can be extracted as features using wavelet packet decomposition.
• Waveform similarity: Using the Dynamic Time Warping (DTW) algorithm to calculate the similarity of each line’s transient waveform to typical fault waveforms, the faulted line shows higher similarity.

4.4 Classification Decision

After extracting features, a classifier is needed to determine which line is the faulted line. Common methods include:

• Threshold method: Setting thresholds for features such as amplitude and polarity (e.g., if the current amplitude of a certain line exceeds twice the average of other lines), exceeding the threshold is determined as the faulted line. This method is simple and fast, suitable for steady-state scenarios.
• Machine learning method: Extracting multi-dimensional features (such as amplitude, phase, high-frequency energy) as input to train classifiers like Support Vector Machines (SVM), Random Forests (RF), or Deep Neural Networks (DNN). This method is more adaptable to complex scenarios (such as multi-branch, high-resistance grounding).
• Pattern matching method: Establishing characteristic templates of typical fault lines (such as the “negative polarity pulse + high-frequency oscillation” pattern of transient waveforms) and judging the faulted line through template matching degree.

5. Key Technologies and Breakthroughs

5.1 Adaptability to Complex Operating Modes

With the integration of distributed generation (DG) and energy storage devices into the distribution network, the traditional single-source radial structure has transformed into a multi-source network, significantly changing the current distribution characteristics:

• The injected current from DG may superimpose on the faulted line, potentially increasing the current amplitude of non-faulted lines and obscuring fault characteristics.
• In dual-source power supply scenarios, fault current may come from two directions, requiring re-modeling of the current distribution path.

Solution: Employing multi-source network topology analysis to establish a node-branch admittance matrix that includes DG nodes, correcting the current distribution model based on the direction information of the injected current (such as the injection point location).

5.2 Feature Extraction for High-Resistance Grounding Faults

In high-resistance grounding (grounding resistance <span>R<span>f</span> > 1000</span><span>Ω</span>), the fault current is weak (usually less than 1A), and the low-frequency signals actively injected may be drowned out by noise, while the amplitude of self-excited transient currents will also significantly decrease.

Solution:

• Using high-sensitivity sensors (such as Rogowski coils) to improve the ability to capture transient signals;
• Utilizing Blind Source Separation (BSS) algorithms (such as Independent Component Analysis, ICA) to separate weak fault components from mixed signals;
• Combining harmonic injection methods (such as injecting third and fifth harmonics) to enhance features using the attenuation characteristics of harmonics in faulted lines.

5.3 Feature Identification in Noisy Environments

There is a significant amount of electromagnetic interference in the distribution network (such as switching operations and lightning), leading to low signal-to-noise ratios (SNR) in the collected current signals (possibly below 10dB), and traditional filtering methods may lose effective features.

Solution:

• Using adaptive filtering algorithms (such as LMS algorithms) to suppress periodic interference in real-time;
• Utilizing the multi-resolution characteristics of wavelet transforms to extract detailed information of fault features at different scales;
• Introducing redundant sensors (such as deploying multiple CTs at key nodes) to improve feature reliability through a majority voting mechanism.

6. Engineering Applications and Validation

Practical tests in a 10kV distribution network (with the neutral point grounded through an arc suppression coil) have validated the effectiveness of this method. The test scenarios include:

• Single-phase high-resistance grounding (<span>R<span>f</span> = 2000</span><code><span>Ω</span>): Actively injecting a 10Hz low-frequency current (amplitude 1A), the 10Hz current amplitude in the faulted line is 0.8A (only 0.1A in the non-faulted line), accurately identifying the faulted line through the threshold method.
• Intermittent arc grounding: Collecting transient current signals, it was found that the faulted line exhibited a negative polarity pulse with an amplitude of 2.5A at the moment of arc reignition (no such feature in the non-faulted line), achieving line selection in conjunction with the polarity reversal criterion.
• Multi-branch lines (5 outgoing lines): After injecting pulsed current, extracting the energy ratio in the 3-10kHz frequency band for each line through wavelet packet decomposition, the energy ratio of the faulted line reached 65% (only 15%-25% for non-faulted lines), with correct classification.

7. Summary and Outlook

The fault line selection method based on injected current distribution characteristics effectively addresses the limitations of traditional methods in scenarios such as high-resistance grounding and arc suppression coil compensation by actively injecting signals or utilizing fault self-excited currents to explore the distribution differences of current in faulted and non-faulted lines. Future research directions include:

• Multi-source information fusion: Combining multi-dimensional information such as voltage and zero-sequence power to enhance the reliability of line selection in complex scenarios;
• Digital twin technology: Establishing a digital twin model of the distribution network to simulate the current distribution characteristics under different fault scenarios in real-time, optimizing feature extraction algorithms;
• Edge computing applications: Deploying lightweight machine learning models at distribution terminals to achieve rapid local identification of fault features, reducing communication latency.

This method provides important technical support for the high reliability and intelligent operation and maintenance of distribution networks, with significant engineering application prospects.