SVM Based Fault Detection, Classification in Distribution System Using MATLAB

Support Vector Machine (SVM) Based Fault Detection and Classification in Electrical Distribution System Using MATLAB/Simulink

Product Description

The SVM-Based Fault Detection and Classification System is a MATLAB/Simulink-based intelligent power system protection model designed to automatically detect and classify electrical faults using supervised machine learning.

The system employs a Support Vector Machine (SVM) algorithm to analyze three-phase voltages, currents, and zero-sequence components to distinguish normal and faulty operating conditions.

The simulation consists of an 11 kV three-phase source, 11 kV/400 V distribution transformer, three-phase load, and configurable fault block.

The SVM classifier is trained using MATLAB Classification Learner with a Fine Gaussian (RBF) kernel and five-fold cross-validation. The demonstrated model achieves approximately 96.67% cross-

validation accuracy.

The trained classifier is integrated into Simulink for automatic fault detection, classification, and visualization.

Technical Specifications

Working Principle

1. Distribution System Modeling

An 11 kV three-phase electrical source supplies a distribution transformer connected to a three-phase load. Configurable fault blocks introduce different short-circuit conditions. Voltage, current, and RMS measurement blocks monitor system performance.

2. Fault Dataset Generation

Electrical measurements are collected under normal and fault conditions.

• Input Dataset (X): Three-phase voltages, currents, and zero-sequence components.

• Target Dataset (D): Corresponding fault classification labels.

These datasets provide training information for the SVM classifier.

3. SVM Training and Validation

The collected datasets are imported into MATLAB Classification Learner.

• Select SVM classification.

• Configure the Fine Gaussian kernel.

• Apply five-fold cross-validation.

• Train and evaluate the classifier.

• Analyze confusion matrix and ROC results.

• Export the trained SVM model.

Demonstrated Classification Accuracy: 96.67%.

4. Simulink Integration

The trained SVM model is connected to the distribution system through an Interpreted MATLAB Function block.

Measured voltage and current signals are supplied to the classifier. Its predicted output is decoded to determine the corresponding fault condition.

5. Fault Detection and Classification

The model supports the following fault categories:

The simulation displays the predicted fault class, fault detection signal, and voltage/current waveform responses.

Key Features

• Automatic SVM-based fault detection and classification.

• Three-phase distribution system simulation.

• Fine Gaussian SVM implementation.

• Approximately 96.67% demonstrated validation accuracy.

• Five-fold cross-validation.

• Multiple symmetrical and unsymmetrical fault categories.

• Voltage, current, and zero-sequence feature extraction.

• MATLAB Classification Learner integration.

• Trained SVM classifier integrated with Simulink.

• Configurable fault simulation.

• Confusion matrix and ROC-based performance evaluation.

• Three-phase RMS and waveform visualization.

• Suitable for power system protection and ML research.

Applications

• Distribution System Protection: Intelligent detection and classification of electrical faults.

• Transmission Line Protection: Extend the methodology using appropriate transmission line training data.

• Smart Grid Applications: Machine learning-based electrical network monitoring.

• Power System Fault Analysis: Investigate symmetrical and unsymmetrical fault characteristics.

• MATLAB Academic Projects: Practical implementation of SVM in electrical engineering.

• Research Development: Compare SVM with ANN, KNN, Decision Tree, and Random Forest algorithms.

Simulation Results

The MATLAB/Simulink model demonstrates:

• Normal operating condition identification.

• Single line-to-ground fault classification.

• Line-to-line and double line-to-ground fault identification.

• Three-phase fault classification.

• Voltage and current waveform analysis.

• RMS voltage and current variations.

• SVM prediction and fault classification outputs.

• Fault detection timing visualization.

• Confusion matrix, ROC curves, and classification accuracy.

The 96.67% accuracy represents the demonstrated five-fold cross-validation performance and may vary with training datasets, fault parameters, and network operating conditions.