Fault Detection, Classification, and Location in Power System using ANFIS
Fault Detection, Classification, and Location in Power System using ANFIS
𝐈𝐧𝐭𝐫𝐨𝐝𝐮𝐜𝐭𝐢𝐨𝐧
Fast and accurate fault identification is essential for improving the reliability, protection, and safety of electrical power systems.
When a fault occurs on a transmission or distribution line, the protection system should determine three important things:
Whether a fault has occurred
What type of fault has occurred
Where the fault is located
In this MATLAB/Simulink model, an Adaptive Neuro-Fuzzy Inference System (ANFIS) is used for fault detection, fault classification, and fault location estimation.
ANFIS combines the learning capability of a neural network with the decision-making capability of a fuzzy inference system. This makes it suitable for nonlinear power-system problems where conventional threshold-based techniques may have difficulty distinguishing between similar fault conditions.
𝐖𝐡𝐚𝐭 𝐢𝐬 𝐀𝐍𝐅𝐈𝐒?
ANFIS stands for Adaptive Neuro-Fuzzy Inference System.
It combines two intelligent techniques:
Neural Network: Learns the relationship between input and output data.
Fuzzy Logic: Uses fuzzy rules and membership functions for decision-making.
After training, the ANFIS model can recognize patterns in measured electrical quantities and generate the required fault information.
In this application, separate ANFIS models are used for:
Fault detection and classification
Fault location estimation
𝐒𝐲𝐬𝐭𝐞𝐦 𝐎𝐯𝐞𝐫𝐯𝐢𝐞𝐰
The MATLAB/Simulink power system consists of source feeders, buses, a distribution line, transformer, load, measurement blocks, and ANFIS-based intelligent fault analysis.
Parameter | Description |
Source Voltage | 11 kV |
Source Rating | 30 MVA |
Number of Main Buses | 3 |
Line Length | 10 km |
Transformer Primary | 11 kV |
Transformer Secondary | 400 V |
Fault Analysis Method | ANFIS |
Main Platform | MATLAB/Simulink |
Bus 1 and Bus 3 are connected through the power line where different fault conditions are introduced.
The transformer reduces the voltage from 11 kV to 400 V, representing the customer-side voltage level.
𝐏𝐨𝐰𝐞𝐫 𝐒𝐲𝐬𝐭𝐞𝐦 𝐅𝐚𝐮𝐥𝐭 𝐓𝐲𝐩𝐞𝐬
The developed model can analyze different single-line, double-line, and three-phase fault conditions.
Typical fault classes considered in the system include:
Fault Code | Fault Type |
AG | Phase A to Ground |
BG | Phase B to Ground |
CG | Phase C to Ground |
AB | Phase A–Phase B |
AC | Phase A–Phase C |
BC | Phase B–Phase C |
ABG | Phase A–Phase B to Ground |
ACG | Phase A–Phase C to Ground |
BCG | Phase B–Phase C to Ground |
ABC | Three-Phase Fault |
ABCG | Three-Phase to Ground Fault |
A normal operating condition is also included in the training dataset.
𝐈𝐧𝐩𝐮𝐭 𝐃𝐚𝐭𝐚 𝐟𝐨𝐫 𝐀𝐍𝐅𝐈𝐒
Electrical signals are measured from the power system during both normal and faulty operating conditions.
Important measured quantities include:
RMS phase voltage
RMS phase current
Zero-sequence voltage
Zero-sequence current
These measurements contain valuable information about the type and severity of a fault.
The fault-classification ANFIS receives the required voltage and current features, while the fault-location ANFIS uses voltage-related information to estimate the distance of the fault from the measurement point.
𝐃𝐚𝐭𝐚𝐬𝐞𝐭 𝐆𝐞𝐧𝐞𝐫𝐚𝐭𝐢𝐨𝐧
Before ANFIS can identify faults online, it must first be trained using representative power-system data.
The simulation is executed for:
Normal conditions
Different fault types
Different fault locations
Different electrical responses
For every simulation case, the necessary voltage and current features are recorded and associated with predefined target outputs.
The accumulated input and target datasets are then used for ANFIS training.
𝐁𝐢𝐧𝐚𝐫𝐲 𝐅𝐚𝐮𝐥𝐭 𝐄𝐧𝐜𝐨𝐝𝐢𝐧𝐠
The classification system generates a four-bit binary output.
The binary information is decoded to determine the corresponding fault condition.
For example:
Operating Condition | Example Target |
Normal Condition | 0000 |
AG Fault | Example: 1001 |
Other Fault Conditions | Assigned individual binary combinations |
The exact target code assigned to each fault class depends on the coding strategy used during dataset preparation.
This binary-output method makes it easier to connect the ANFIS classifier to Simulink logic blocks for online fault identification.
𝐀𝐍𝐅𝐈𝐒 𝐓𝐫𝐚𝐢𝐧𝐢𝐧𝐠 𝐒𝐭𝐫𝐚𝐭𝐞𝐠𝐲
Two separate intelligent systems are developed.
ANFIS Model | Purpose | Training Approach |
ANFIS 1 | Fault Detection and Classification | Subtractive Clustering |
ANFIS 2 | Fault Location Estimation | Grid Partition |
Fault Classification ANFIS
The classification ANFIS is trained using measured power-system parameters.
Its main responsibilities are to:
Detect the presence of a fault
Identify the corresponding fault category
Generate the required binary output
Fault Location ANFIS
The second ANFIS model estimates the distance between the measurement point and the fault.
Its output is directly interpreted as the estimated fault location in kilometers.
𝐖𝐨𝐫𝐤𝐢𝐧𝐠 𝐏𝐫𝐨𝐜𝐞𝐬𝐬
The complete fault-analysis procedure can be summarized as:
The power system operates under normal conditions.
A fault is introduced on the line.
Bus voltage and current signals change immediately.
RMS and zero-sequence quantities are extracted.
The measured features are supplied to the trained ANFIS models.
The classification ANFIS detects the fault.
The classifier determines the fault category.
The location ANFIS estimates the distance of the fault.
Binary outputs are decoded in Simulink.
Fault type and estimated location are displayed.
This entire process can be performed during simulation without manually analyzing the waveform after every fault.
𝐅𝐚𝐮𝐥𝐭 𝐃𝐞𝐭𝐞𝐜𝐭𝐢𝐨𝐧 𝐚𝐧𝐝 𝐂𝐥𝐚𝐬𝐬𝐢𝐟𝐢𝐜𝐚𝐭𝐢𝐨𝐧
Once the fault occurs, the ANFIS classifier examines the measured electrical parameters.
For example, when an AG fault is applied:
The fault detector changes its output.
The AG classification output becomes active.
Other fault-classification outputs remain inactive.
The location model simultaneously estimates the fault distance.
Therefore, detection, classification, and location can be obtained from the same power-system simulation.
𝐅𝐚𝐮𝐥𝐭 𝐋𝐨𝐜𝐚𝐭𝐢𝐨𝐧 𝐄𝐬𝐭𝐢𝐦𝐚𝐭𝐢𝐨𝐧
The location ANFIS is trained using datasets collected at different fault positions.
During online simulation, the trained model compares the new measured pattern with the learned patterns and estimates the corresponding location.
The simulation demonstrates that the estimated location remains reasonably close to the actual fault position for the tested cases.
𝐒𝐢𝐦𝐮𝐥𝐚𝐭𝐢𝐨𝐧 𝐑𝐞𝐬𝐮𝐥𝐭𝐬
Several fault conditions were tested after completing the ANFIS training.
Selected Fault-Location Results
Fault Type | Actual Location | Estimated Location | Classification |
ABCG Fault | 6 km | 6.42 km | Correct |
Fault at 2 km | 2 km | 2.43 km | Correct |
AG Fault | 2 km | 2.10 km | Correct |
ABC Fault | 2 km | 1.90 km | Correct |
BC Fault | 2 km | 2.015 km | Correct |
ABCG Fault | 2 km | 1.90 km | Detected, but classified as ABC |
These results show that the ANFIS location model can provide a close estimation of the actual fault position.
𝐂𝐥𝐚𝐬𝐬𝐢𝐟𝐢𝐜𝐚𝐭𝐢𝐨𝐧 𝐏𝐞𝐫𝐟𝐨𝐫𝐦𝐚𝐧𝐜𝐞
During the demonstrated tests:
Performance Indicator | Result |
Number of Fault Types | 11 |
Correctly Classified | 10 |
Misclassified | 1 |
Approximate Classification Accuracy | 91% |
The primary misclassification occurred between ABC and ABCG faults.
This can happen because their voltage and current patterns may be very similar under certain operating conditions.
Additional training samples and carefully selected input features can help improve the distinction between such closely related fault types.
𝐀𝐍𝐅𝐈𝐒 𝐅𝐮𝐳𝐳𝐲 𝐑𝐮𝐥𝐞 𝐒𝐲𝐬𝐭𝐞𝐦
After training, MATLAB generates fuzzy membership functions and rules based on the dataset.
When the measured inputs change:
Membership values change.
Different fuzzy rules become active.
ANFIS calculates the corresponding output.
The resulting output identifies the fault or its location.
This provides a data-driven intelligent alternative to purely fixed threshold-based fault identification.
𝐖𝐡𝐲 𝐔𝐬𝐞 𝐙𝐞𝐫𝐨-𝐒𝐞𝐪𝐮𝐞𝐧𝐜𝐞 𝐐𝐮𝐚𝐧𝐭𝐢𝐭𝐢𝐞𝐬?
Zero-sequence voltage and current are particularly useful when analyzing ground-related faults.
They help the intelligent classifier distinguish conditions involving a ground connection from other phase-to-phase faults.
Combining:
Phase measurements
RMS quantities
Zero-sequence components
provides the ANFIS classifier with more information about the fault condition.
𝐊𝐞𝐲 𝐅𝐞𝐚𝐭𝐮𝐫𝐞𝐬
ANFIS-based intelligent fault analysis
Simultaneous fault detection, classification, and location
MATLAB/Simulink implementation
Supports multiple power-system fault categories
Uses actual electrical measurements as ANFIS inputs
Includes RMS voltage and RMS current information
Includes zero-sequence voltage and current information
Separate ANFIS models for classification and location
Binary-coded classification output
Online visualization of fault status
Fault distance estimation in kilometers
Approximately 91% classification accuracy in the demonstrated test set
Suitable for studying intelligent protection techniques
𝐁𝐞𝐧𝐞𝐟𝐢𝐭𝐬 𝐨𝐟 𝐀𝐍𝐅𝐈𝐒-𝐁𝐚𝐬𝐞𝐝 𝐅𝐚𝐮𝐥𝐭 𝐀𝐧𝐚𝐥𝐲𝐬𝐢𝐬
The ANFIS approach provides several advantages for power-system fault studies.
Learns nonlinear relationships from simulation data
Combines neural learning and fuzzy decision-making
Can distinguish multiple fault categories
Provides numerical fault-location estimates
Can be retrained using additional operating conditions
Suitable for integrating with MATLAB/Simulink protection models
Reduces the need for manual waveform interpretation
𝐀𝐩𝐩𝐥𝐢𝐜𝐚𝐭𝐢𝐨𝐧𝐬
ANFIS-based fault detection techniques can be studied for applications such as:
Transmission-line protection
Distribution-system monitoring
Intelligent protection schemes
Substation monitoring
Fault-location systems
Smart-grid protection
Microgrid monitoring
Renewable-energy integrated networks
Automated fault diagnosis
Power-system condition monitoring
𝐖𝐡𝐨 𝐂𝐚𝐧 𝐔𝐬𝐞 𝐓𝐡𝐢𝐬 𝐌𝐨𝐝𝐞𝐥?
This simulation is useful for:
Electrical engineering students
Power-system researchers
MATLAB/Simulink learners
Protection engineers
Smart-grid researchers
Engineers studying artificial intelligence in power systems
Researchers working with ANFIS, neural networks, and fuzzy logic
𝐖𝐡𝐚𝐭 𝐂𝐚𝐧 𝐁𝐞 𝐋𝐞𝐚𝐫𝐧𝐞𝐝?
By studying this model, users can understand:
How faults are created in a power-system model
How fault data are collected
How RMS quantities are used for fault analysis
How zero-sequence quantities help identify ground faults
How training datasets are prepared
How ANFIS is trained in MATLAB
How fuzzy rules are generated
How faults can be represented using binary outputs
How fault location can be estimated using intelligent algorithms
How trained ANFIS models are integrated into Simulink
𝐂𝐨𝐧𝐜𝐥𝐮𝐬𝐢𝐨𝐧
The Fault Detection, Classification, and Location in Power System using ANFIS model demonstrates how artificial-intelligence techniques can be integrated with MATLAB/Simulink for intelligent power-system protection.
The developed system uses measured voltage and current information to determine the presence and category of a fault while a separate ANFIS model estimates its location.
In the demonstrated cases, 10 out of 11 fault types were correctly classified, corresponding to approximately 91% classification accuracy. Fault-location estimates such as 6.42 km for an actual 6 km fault and 2.015 km for an actual 2 km fault also demonstrate the ability of the trained ANFIS model to estimate fault distance effectively.
The model provides a clear learning platform for understanding ANFIS, fault diagnosis, intelligent protection, fuzzy logic, and power-system monitoring in MATLAB/Simulink.
Comments