Neural Network for FCEVs and RM Power Management using V2G technology

this study investigates the effectiveness of various training algorithms for engineering simulations and integrates the Fuel Cell Electric Vehicle to Grid system into its energy management objectives. The system, linking fuel cell electric vehicles to the grid, is examined alongside neural fitting techniques using MATLAB R2019a version. The Levenberg-Marquardt algorithm stands out with a training error of 0.00073 and a correlation rate of 99.907%. During validation and testing, it maintains high correlation rates of $99.917 \%$ and $99.907 \%$, with minimal errors of 0.0013 and 0.0007, respectively. Bayesian Regularization shows promise with a training error of +0.00092 and a correlation rate of 99.884%, but concerns arise due to a marginal test correlation rate of 99.618% and a negative error of -0.000012. In contrast, the Scaled Conjugate Gradient algorithm performs less favourably, with a training error of +0.016 and a correlation rate of $98.064 \%$. Validation rates drop significantly to $9.129 \%$, with an error of 0.012, and testing reveals a correlation rate of 97.193% with an error of 0.28. The Levenberg-Marquardt algorithm demonstrates supremacy in neural fitting, with a remarkable correlation rate of 99.909%. Although Bayesian Regularization shows lower error values, Levenberg-Marquardt’s validation metrics reinforce its dominance, with a best performance of 0.00081724 for mean squared error. In summary, the Levenberg-Marquardt algorithm emerges as highly effective for energy management simulations, offering superior correlation rates and minimal errors compared to Bayesian Regularization and Scaled Conjugate Gradient.