A Lightweight Sequential Convolutional Neural Network for Smart Grid Stability Analysis

Abstract

A Smart grid stability analysis is essential for ensuring modern power systems' reliable and secure operation. This approach helps identify potential instabilities and disturbances that can lead to blackouts or equipment failures. By analyzing the stability of the grid, operators can take proactive measures to maintain a stable and resilient power infrastructure. Monitoring smart grid data from various sources and analyzing how to control the stability of the grid are challenging tasks. A convolutional neural network (CNN) can effectively capture spatial dependencies and patterns from grid data and can help in enabling accurate prediction and classification of stability-related events in a power system. However, developing a CNN that has fewer learnable parameters and provides high accuracy is challenging. This paper presents a sequential CNN architecture to detect the stability of the Decentral Smart Grid Control (DSGC) system. A mathematical model of the 4-node start architecture of a smart grid was presented. Later, 12 parameter-based grid datasets from the UCI repository were used to validate the proposed network. The proposed CNN accepts sequential data to capture temporal dependencies in the data. The sequential process in a single dimension offers fewer learnable parameters, making the network more compact and computationally efficient. The proposed 11-layered CNN has a total of 12.7K learnable parameters. The detailed analysis of the proposed CNN using ambiguity and the t-SNE score suggested that the model can identify discriminative features for classifying data into stable and unstable classes. A comparison analysis of the quantitative parameters revealed that the model performed well, with 98.82%, 98.55%, 98.88%, and 98.77% accuracy, precision, recall, and F1 score, respectively.