A Hybrid Bio-Inspired GRU Model with Multiheaded Attention for Predicting Smart Grid Stability

Abstract

Smart grids are a big part of modern energy systems because they make it easier to handle and distribute energy. Still, it's hard to keep the grid stable because working conditions, supply, and demand for energy are always changing. The Bioinspired (PSO) + GRU Model is a strong way to handle the complexity of grid stability prediction that is shown in this work. An algorithm based on biology and deep learning make the system very accurate and reliable. It uses Gated Recurrent Units (GRUs) to model time and Particle Swarm Optimisation (PSO) to choose which features to use. The suggested design uses the Smart Grid Stability collection, which has 60,000 records of grid properties such as delay, flexibility, and power. It also has a goal variable that shows how stable the grid is. Using tools like SMote can help even out the classes in a dataset, which makes sure that the model is trained fairly. PSO is used to pick the most important features, which lowers the number of dimensions and speeds up processing. GRUs can find sequential links in the grid's operational data using Multiheaded attention methods. This lets them make accurate predictions about stability. For example, SVC, LGBR, and ANN are not as accurate as this model, which has an F1-score of 99.1% and an accuracy rate of 99.5%. The confusion matrix study shows that the framework is even more stable, with low judgement mistakes. Thorough planning steps, such as normalisation, dimensionality reduction, data cleaning, and more, make sure that the model gets good inputs. This paper stresses how important it is for smart grid systems to use cutting edge deep learning and efficiency methods to solve real-world problems. The suggested method could be used to make predictions more accurate and could also be scaled up to be used in real-time grid tracking systems. In the future, this method could be used in other study areas that need model time data and big decisions.