Integrating Distributed Generation and Advanced Deep Learning for Efficient Distribution System Management and Fault Detection

Distribution system voltage profile management at each bus and fault detection and classification are often challenged by complex and changing network configurations. The distribution system voltage profile improvement issue is addressed by placing distributed generation (DG) units at different locations in the network. By placing the DG units at appropriate places in IEEE 33 bus radial distribution networks by a proposed reinforcement learning (RL) algorithm, the voltage profile of each node is improved and power loss in the network is minimized. There is a 69% reduction in active power losses compared to losses without DG. Furthermore, an innovative method for fault detection and classification is developed that uses a convolutional neural network (CNN) cascaded with a long short-term memory network (LSTM) and attention mechanisms (AMs). To extract dynamic information from the data, phasor measurement units (PMUs) placed on different buses are used as input for the CNN architecture. AM strengthens important information. A mapping weight and parameter learning approach allows AM to assign different weights to concentrate on LSTM characteristics and improve learning accuracy. Low and high impedance faults are tested as well as various non-faulty events. The scheme's performance is compared with that of other deep learning techniques through reliability analysis, and the time taken for fault detection (FD) is also determined.