Low computational cost convolutional neural network for smart grid frequency stability prediction

Abstract

In the smart grid, it is critical to collect dynamic and time-dependent information on energy demand and consumption and compare it to current supply conditions. The decentral smart grid control (DSGC) system manages frequency, a Smart grid element. It connects energy costs to grid frequency, allowing access to both consumers and producers. This work proposes a pruning of the convolution layers and neurons of 1-dimensional time-aware convolutional neural network (1D CNN) analysis of grid frequency stability to determine efficient energy costs. The proposed solution evaluated augmented grid stability datasets in addition to two other publicly available datasets to ascertain the approach’s feasibility in various scenarios; the simulation demonstrated a minimal train and prediction time of 124.37 s and 17.67 s efficiency over compared models, with prediction accuracy of 99.79% and 0.01 MFLOPs. Matthew’s correlation coefficient was applied to evaluate further the performance of the proposed 1D CNN to ascertain its applicability in various scenarios.