A Unique Approach to Fault Detection in Power Systems

Dr. Manoj Kumar Nigam

Professor & Head

Department of Electrical Engineering

Kalinga University, Raipur (C.G)

manoj.nigam@kalingauniversity.ac.in

Transmission lines (TL) are in charge of transferring power across the grid in electrical power systems. However, faults in these lines are abnormal conditions that, if prolonged, can destabilize the transmission system. IEC61850-based digital substations provide sampled value measurements in the substation to examine faults. (ML) techniques are being investigated in addition to existing model-based techniques for fault diagnosis. We we’re studying an ML-based fault classifier that takes current and voltage measurements as input. When compared to other methods proposed in the literature, such as RNN and SVM, the ML-based fault classifier is intended to improve performance in fault classification.

A smart grid with bidirectional data and energy flow is replacing the current power grid. The objective of a stable and dependable grid operation is crucial in a smart grid due to growing generations and demand over time. The smart grid’s transmission system links consumers to a diversity of energy sources, with renewable energy sources. The smart grid cannot operate properly unless the transmission system is safeguarded.

However, natural disasters, severe weather, and human intervention can all result in abnormal conditions. Using a change of protective relays, transmission line protection systems carry out fault analysis, which contains fault detection, classification, and location activities.

The objective of fault diagnosis in the online implementation is to examine the fault in terms of its kind, location, and cause—that is, system disturbances. To provide this fault diagnosis task for the transmission line in digital substations, data-driven approaches have been explored. Classical algorithms and contemporary methodologies are used in TL protection systems to identify and categorize faults.

Fault detection in transmission lines is established through the use of over-current relays, distance relays, and directional relays, each with distinct characteristics. The fault classification is traditionally done using sequence component distance relays to classify the fault in the system using positive, negative, and zero sequence components. Particularly for fault classification, sequence component-based relays are used; however, with the addition of distributed energy generations, the effectiveness of sequence component-based approaches is declining.

Two reasons for using sequence models. First, deep learning (DL) networks are the best at extracting features from temporal data sequentially. Second, unlike RNNs, it overcomes the vanishing gradient problem when learning from sequential data. Another candidate approach was the transformer model, but it does not provide learning in a sequential manner, which is required in power system measurement data. For example, using Long Short-Term Memory (LSTM) networks, temporal information from sequential data can be learned and classified for various abnormal behaviors, including faults. The fault detection and classification task is carried out using input current and voltage measurement data from the substation. The approach of sequence learning models is used to solve the fault classification problem