Deep Neural Network-Based Fault Detection and Classification in Power Systems

Deep Neural Network-Based Fault Detection and Classification in Power Systems

Introduction to Fault Detection and Classification

Fault detection and classification play a significant role in maintaining the safety and reliability of power systems. This process helps in identifying faults that occur within the power grid, thus enabling timely maintenance and avoiding potential damage. In the video tutorial, we explore how to use deep neural networks to automate this fault detection and classification process, using MATLAB for implementation.

Setting Up the System Model

The system model used in the demonstration consists of:

• 11 kV to 400 V Feeder: The feeder steps down high voltage (11 kV) to a lower voltage (400 V) for consumer usage.

• Load Feeder and Transformers: These components manage the distribution of electrical power.

• Fault Creation: Faults are introduced at various points, such as line-to-line and line-to-ground faults, to simulate real-world scenarios.

Data collected includes RMS voltage, RMS current, and sequence currents for both normal and faulty conditions. This data serves as the input for training the neural network.

Advanced Control and Energy System +

Book Now

Fault Detection with Neural Networks

1. Data Collection: For each type of fault, data is collected based on RMS voltage, RMS current, zero-sequence voltage, and zero-sequence current.

2. Simulation: Various fault scenarios are simulated, including line-to-ground, double line-to-ground, and three-phase faults, with different fault resistances.

3. Training the Neural Network: The collected data is used to train a neural network model. Normal operating conditions are marked as "no fault" (output = 0), while faulty conditions are marked as "fault detected" (output = 1). The model is trained to differentiate between these conditions.

Fault Classification

The fault classification model follows a similar setup but focuses on categorizing faults rather than merely detecting their presence. The process involves:

1. Creating Fault Models: Different fault models are created to represent various fault types, such as single-phase to ground or double-phase faults.

2. Data Collection and Target Assignment: For each fault type, data is collected and associated with a four-bit binary target value representing different fault categories.

3. Training the Classification Network: The neural network is trained with the collected data, and the performance is assessed based on how accurately it classifies the faults.

Results and Performance

The tutorial demonstrates that:

• Fault Detection: The neural network model achieves 100% accuracy in detecting faults, indicating that it reliably identifies when a fault occurs.

• Fault Classification: The classification model has a high accuracy rate of approximately 91%. However, it struggled to classify one specific fault type (ABC ground fault) accurately.

The results underscore the effectiveness of using deep neural networks for fault detection and classification in power systems, with minor limitations in fault classification.

Conclusion

Using deep neural networks for fault detection and classification provides a robust solution for managing power system faults. The method allows for precise and automated fault identification and categorization, which enhances system reliability and maintenance efficiency. For those interested in implementing these techniques, MATLAB provides a practical platform for model creation and simulation.