Fault Diagnosis in Smart Grids Using a Deep Long Short-Term Memory-based Feature Learning Architecture
30th European Safety and Reliability Conference, ESREL 2020 and 15th Probabilistic Safety Assessment and Management Conference, PSAM 2020
Classification algorithms, Dimensionality reduction, Fault diagnosis, Recurrent neural networks, Smart grids
This paper deals with the fault cause identification problem in smart grids through a data-driven diagnostic architecture. The proposed approach is based on the use of voltage measurements collected by sparse datarecording devices in normal and abnormal operating conditions of the power grid. A signal processing module firstly refines the voltage measurements which are then fed into a long short-term memory recurrent neural networks to diagnose the state of the system. Furthermore, a comparative study has been fulfilled by modifying echo state networks to deal with multiclass classification problems through a one-versus-all scheme and by contributing the application of features extraction techniques such as principal component analysis, linear discriminant analysis and multidimensional scaling. Various scenarios are simulated on the IEEE 39-bus system for diagnosing faults in normal, noisy, low sampling-rate, and high fault-resistance conditions. The attained results validate the applicability of the proposed framework and denote the superiority of the long short-term memory networks over echo state networks in designed scenarios.
Hassani, Hossein; Razavi-Far, Roozbeh; and Saif, Mehrdad. (2020). Fault Diagnosis in Smart Grids Using a Deep Long Short-Term Memory-based Feature Learning Architecture. 30th European Safety and Reliability Conference, ESREL 2020 and 15th Probabilistic Safety Assessment and Management Conference, PSAM 2020, 1323-1330.