Robust Optical Power Forecasting Across Temperatures: A New L-I Model with Enhanced Temperature Dependence

This study presents a new L-I Model, an optimized optical power prediction model that incorporates temperature-dependent efficiency $(\boldsymbol{\eta}(\mathbf{T}))$, linear temperature-dependent thermal resistance ($\boldsymbol{R}_{\boldsymbol{t h}}$). ReLU activation is applied to ensure non-negative predictions. Using the RMSprop optimizer with a small learning rate (2e-12), the model achieves an $\mathbf{R}^{\mathbf{2}}$ of $\mathbf{0. 6 9 9 2}$, MSE of 0.2004 and MAE of 0.3826 significantly outperforming previous models. The model demonstrates robust performance in predicting optical power across a range of temperatures $\left(10^{\circ} \mathrm{C}\right.$ to $90^{\circ} \mathrm{C}$). Key improvements include enhanced temperature dependence modeling and ReLU activation, resulting in more accurate predictions and stable training.