Ground Current Prediction Methodology Employing RIME-CNN-LSTM-Attention Architecture

Abstract

The secure and dependable functioning of high-voltage cable networks relies on the accurate prediction of ground currents. This research introduces a groundbreaking prediction framework, termed RIME-CNN-LSTM-Attention, which leverages the Rime optimization algorithm to enhance the architecture of Long Short-Term Memory networks (LSTM). The framework integrates Convolutional Neural Networks (CNN) and attention mechanisms to establish a robust predictive backbone. The adaptability of the RIME algorithm is instrumental in optimizing critical hyperparameters such as learning rate, hidden layer dimensions, and regularization coefficients, thereby enhancing the model's global optimization capabilities and reducing the likelihood of converging to suboptimal local optima. The main goal of this research is to create a prediction model for ground currents with a high degree of accuracy, which is crucial for ensuring the secure monitoring and upkeep of high-voltage cable networks. Empirical results validate the exceptional performance of the RIME-CNN-LSTM-Attention model, demonstrating significant reductions in Root Mean Square Error (RMSE) by 57.12%, 51.90%, and 39.95% compared to CNN, LSTM, and CNN-LSTM-Attention models, respectively. This novel approach not only provides robust technical support for the management and maintenance of high-voltage cable systems but also paves new pathways for time-series forecasting research. The study underscores the model's superior predictive performance and robustness, highlighting its substantial academic and engineering significance.