Description:FAULT DETECTION AND CLASSIFICATION IN
TRANSMISSION LINES USING DGR-ELM
Field of the Invention:
The present invention falls within the domain of electrical power systems,
particularly focusing on high-voltage transmission networks. More specifically, it
concerns intelligent fault detection and classification techniques using machine
learning approaches. The core of this invention is built around the DGR-ELM
(Distributed Generalized Regularized - Extreme Learning Machine) algorithm,
which offers a fast, accurate, and scalable solution for identifying faults in
transmission lines.
Background:
Transmission lines are vital for transporting electrical energy over long distances
from generation stations to distribution centers. However, these lines are prone to
faults due to natural events like lightning strikes, equipment degradation, insulation
failure, and environmental influences such as wind or vegetation contact.
Traditional protection schemes, such as overcurrent and differential protection
systems, have limitations. They typically rely on real-time voltage and current
measurements, which can be affected by external noise, electromagnetic
interference (EMI), or sensor inaccuracies. These conventional systems often
struggle with timely and precise fault classification, leading to delayed responses
or unnecessary disconnections.
As modern power grids become increasingly complex and demand higher
reliability, the need for rapid and intelligent fault diagnosis tools has intensified.
This invention proposes the use of the DGR-ELM algorithm, a machine learningbased model that enables fast training and robust performance. It improves
classification accuracy and operational response times, ultimately enhancing grid
stability and minimizing power outages. The approach allows for more adaptive
and intelligent protection mechanisms in modern power systems, marking a
significant advancement over conventional techniques.
In addition, the proposed system is capable of learning from vast and dynamic
datasets, adapting to new fault patterns over time. It supports real-time
implementation with minimal computational overhead, making it suitable for
integration into existing grid infrastructures. This invention paves the way for more
resilient, data-driven power system protection.
Existing System and Its Drawbacks
Traditional fault detection systems show in the figure 1 which are electrical grids
primarily rely on hardware-based techniques such as overcurrent protection,
differential relays, and impedance relays. These systems operate by monitoring
voltage and current levels to detect and isolate faults. For instance, overcurrent
protection triggers a trip mechanism when the current exceeds a predefined
threshold, aiming to prevent damage to equipment. Differential protection works
by comparing the current entering and leaving a specific section of the network—
any significant difference is interpreted as a fault. Similarly, impedance relays use
the relationship between voltage and current to determine the impedance of a line
section, identifying faults based on deviation from normal impedance values.
While these systems have been foundational for decades, they exhibit several
critical drawbacks. Firstly, they are inherently slow in responding to faults,
especially in complex or large-scale grids. They often struggle to distinguish
between permanent faults and transient conditions, such as temporary overloads
or switching events. Furthermore, these devices are vulnerable to external
disturbances, including temperature fluctuations, electromagnetic interference
(EMI), and aging of components, all of which may lead to false positives or
missed detections.
A major limitation is their lack of adaptability—traditional relays cannot learn or
evolve with the system. As modern power grids become increasingly dynamic and
integrated with renewable energy sources, these fixed-rule systems prove
insufficient. They lack the intelligence and scalability needed to handle real-time,
data-driven decision-making.
Fig 1 Existing System block diagram
Summary of the Invention
This invention introduces a cutting-edge approach to electrical fault detection and
classification in power systems using a machine learning technique known as the
Dynamic Generalized Regression Extreme Learning Machine (DGR-ELM) show i
the figure 3. The core innovation lies in its ability to process voltage and current
signals in real time and classify various fault types with high speed and accuracy.
Unlike conventional fault detection systems that rely on complex parameter tuning,
time-consuming training processes, or manual feature engineering, the DGR-ELM
model operates with fixed hidden layer weights show in the figure 2. This design
allows the model to compute output weights analytically, drastically reducing
training time and making the model highly efficient for real-time applications.
Once trained, the model can instantly classify incoming signal patterns into distinct
fault categories or identify a no-fault condition.
The invention significantly outperforms traditional methods in several key areas:
 Rapid Fault Detection: The analytical training process and fast
computation speed allow for near-instantaneous classification of fault types
once data is collected.
 High Classification Accuracy: The model can distinguish between ten
different types of faults and a no-fault condition with an accuracy exceeding
99%.
 Minimal Parameter Tuning: DGR-ELM requires fewer hyper parameters
than deep learning alternatives, simplifying deployment and maintenance.
 Real-Time Monitoring: The system is suitable for continuous, real-time
operation, making it ideal for integration into modern supervisory control
and data acquisition (SCADA) systems and smart grid infrastructures.
 Predictive Capabilities: By utilizing historical data and feedback learning,
the model can anticipate fault trends, providing predictive insights in
addition to real-time diagnosis.
Furthermore, the invention enhances grid reliability by enabling faster fault
resolution and reducing system downtime. Operators benefit from immediate and
accurate diagnostic information, allowing for prompt corrective actions. This level
of operational intelligence is crucial for maintaining system stability, especially as
power networks become more complex with the integration of renewable energy
sources and distributed generation.
The invention’s compatibility with existing smart grid and SCADA technologies
ensures ease of implementation, while its scalable design supports deployment in a
wide range of environments, from small substations to national power networks.
Overall, the proposed DGR-ELM-based fault classification system represents a
significant advancement in the field of intelligent grid management, offering a
reliable, efficient, and forward-looking solution for power system monitoring and
fault analysis.
System Overview and Components
The fault detection system consists of the following key components:
1. Sensors: Measure three-phase voltages and currents in real time from
transmission lines.
2. Preprocessing Unit: Applies min-max normalization to scale the data.
3. DGR-ELM Model: Trained on thousands of data samples, it classifies
inputs into fault types or a no-fault state.
4. Alert System: Triggers warnings and maintenance suggestions upon
detecting a fault.
5. User Interface: Displays real-time status and waveforms.
The architecture is designed to be modular and scalable, making it easy to deploy
across various grid configurations. The machine learning model requires no
complex backpropagation, significantly reducing computational overhead. Because
it uses a single hidden layer with a fixed number of nodes, the system is both light
Fig 2 Proposed framework for fault detection and fault classification
Fig 3 DGR-ELM ARCHITECTURE
Simulation Details and Data Generation
To validate the effectiveness of the proposed fault detection system, extensive
simulations were conducted using MATLAB Simulink, a robust platform for
modeling dynamic power systems. Two distinct transmission line configurations
were developed to mimic real-world scenarios and ensure model adaptability
across different grid setups.
Transmission Line Configurations
1. TL1 – Basic Line Model show in the figure 4
o Components: One synchronous generator, one RLC (Resistor–
Inductor–Capacitor) load
o Line Length: 100 km
o Purpose: This setup serves as a baseline for single-source power
delivery, commonly found in smaller or radial networks.
2. TL2 – Complex Line Model
o Components: Two synchronous generators, three RLC loads
o Line Length: 100 km
o Purpose: Designed to simulate a more complex and meshed network,
this model reflects power sharing, load balancing, and the fault
behavior of interconnected grids.
Fault Injection Strategy
To train the DGR-ELM model effectively, faults are introduced under controlled
conditions:
 Fault Types:
o Single Line-to-Ground (SLG): AG, BG, CG
o Double Line-to-Ground (DLG): ABG, BCG, ACG
o Line-to-Line (LL): AB, BC, AC
o Three-Phase: ABC, ABCG
 Fault Parameters:
o Locations: Randomized along the 100 km line at varying percentages
(e.g., 10%, 30%, 70%, etc.)
o Times: Injected at different moments during the simulation window to
capture both early and late fault dynamics
o Duration: Configurable fault durations to replicate transient and
permanent faults
Each scenario is simulated multiple times with different random seeds to
generate diverse and robust datasets.
Data Collection and Processing
 Waveform Acquisition:
For each simulation run, per-unit (p.u.) voltage and current waveforms for
all three phases are recorded. These per-unit values normalize the electrical
quantities against base values, facilitating consistent analysis across varying
system configurations.
 Data Augmentation:
To improve generalization and robustness, realistic noise and signal
distortion are introduced, replicating the non-ideal conditions present in
practical power systems.
 Labeling and Normalization:
All waveform data is labeled according to the type and location of the fault.
Before feeding into the model, the datasets are normalized using min-max
scaling to maintain consistency in amplitude ranges and reduce numerical
bias during training.
 Dataset Size:
The entire simulation campaign yields thousands of labeled samples, each
containing rich temporal and spatial fault information. These are split into
training, validation, and testing sets for model evaluation.
Realism and Reliability
All simulations are performed using realistic grid parameters, such as line
impedance, generator ratings, and load characteristics. The models incorporate:
 Line capacitance and inductive effects
 Source impedance
 Noise modeling
 Measurement error simulation (e.g., slight phase delays, harmonics)
This level of detail ensures that the DGR-ELM model trained on this data will
perform reliably in real-time, real-world deployment scenarios.
Fig 4 MATLAB Simulink simulation model for 1st transmission line (TL1).
Fig 5 MATLAB Simulink simulation model for 2nd transmission line
(TL2).
Voltage & Current Analysis + Fault Signatures
In fault conditions, the system records significant deviations in voltage and current
waveforms. For example, in an ABG fault:
 Voltage in phases A and B drops suddenly to near zero
 The current rises sharply due to short-circuit behavior
These signal distortions form the basis for classification. The system converts each
waveform into feature vectors containing voltage and current for all three phases.
Normalization is applied using:
Normalized Data = X − MIN(X) MAX(X) − MIN(X)
This ensures data across different scales are comparable, reducing training noise
and convergence errors.
Fig 6 Voltage wave shape during the fault (ABG).
Fig 7 Current wave shape during the fault (ABG).
Claims
Claim 1:
A fault detection system using DGR-ELM for analyzing voltage and current
signals in electrical transmission lines.
Claim 2:
The system of claim 1, wherein a single hidden-layer neural network is used with
fixed random hidden weights and analytically computed output weights.
Claim 3:
The system of claim 1, wherein the signals are normalized using min-max scaling
to ensure uniformity in feature representation.
Claim 4:
The system of claim 1, capable of classifying at least 10 different types of faults
including single-phase, double-phase, and three-phase faults.
Claim 5:
The system of claim 1, which achieves fault classification accuracy exceeding 99%
in simulations and is deployable in real-time environments.
Abstract
Faults in transmission lines pose serious risks to power system stability, often
leading to outages and financial losses. Rapid and accurate fault detection (FD) and
fault classification (FC) are essential to mitigate these issues. This invention
presents a machine learning-based approach using the Distributed Generalized
Regularized - Extreme Learning Machine (DGR-ELM) for efficient FD and FC.
Transmission line behavior under normal and fault conditions was simulated in
MATLAB Simulink, generating voltage and current data across various scenarios.
The system leverages an enhanced version of DGR-ELM, incorporating distributed
computing and regularization techniques to boost performance and scalability.
Compared to traditional Artificial Neural Networks (ANN), DGR-ELM offers
lower computational complexity, faster processing, and minimal parameter tuning.
This enables high-speed, real-time fault classification, improving grid reliability.
The system's capability to process large datasets with high accuracy reduces outage
durations, offering a cost-effective and robust solution for modern power system
fault management. , Claims:Claim 1:
A real-time fault detection and classification system for high-voltage transmission lines, utilizing the DGR-ELM (Distributed Generalized Regularized - Extreme Learning Machine) algorithm, capable of rapidly identifying and classifying fault types with high accuracy, without the need for manual parameter adjustments.
Claim 2:
A low-computational machine learning model for power system fault detection, designed with fixed hidden-layer weights and analytically computed output weights, enabling quick training and efficient real-time operation.
Claim 3:
An adaptive learning mechanism within the DGR-ELM-based fault detection system, capable of continuously learning from historical grid data to recognize emerging fault patterns, enhancing prediction accuracy and grid reliability.