Design of an Improved Model for Electrical Circuit Analysis Using LSTM, BNN, and SR-GP with Laplacian Operations

Differential equations that correctly define circuit behavior are at the very core of Electrical Engineering in the prediction and achievement of optimal performance. Traditional methods usually fail due to the complexities and dynamic nature of modern electrical circuits. This paper will therefore help bridge these shortcomings by applying advanced computational techniques in the analysis and solution for differential equations in electrical circuits, with special reference to RLC circuits. Existing methods for solution of differential equations in electrical circuits often suffer from non-linearity, uncertainty of parameters, and the necessity of making predictions in real time. Conventional techniques have low robustness and may well not lead to full understanding of how circuits evolve over time. A unified framework bringing together LSTM, Bayesian Neural Networks, and Genetic Programming for Symbolic Regression is presented in this paper. The proposed model uses LSTM networks for predictive modeling, tending to leverage their ability in capturing long-term dependencies within time-series data samples. In this manner, a highly accurate voltage and current prediction of about 95% accuracy over a 10step prediction horizon could be achieved.