This study proposes a temperature compensation framework for monitoring the refractive index of lead-acid battery electrolytes. We combine parallel Fabry-Perot Fiber Bragg grating (FP-FBG) sensors with deep learning-based demodulation techniques to effectively suppress thermal interference effects. We innovatively construct a one-dimensional residual convolutional neural network (1D-ResCNN) architecture that directly processes raw two-dimensional spectral data for electrolyte densities ranging from 0.991 g/cm3 to 1.4687 g/cm3 at temperatures between 20 ℃ and 45 ℃, eliminating the need for traditional preprocessing procedures. Experimental results demonstrate that the model converges at the 34th epoch, achieving a mean absolute error of 0.0003 on the validation set with a coefficient of determination of 0.9878, significantly outperforming conventional regression methods. Through dynamic learning rate optimization, the total training time is reduced to just 20 min. The model completes predictions for 360 independent test samples within 3 s, demonstrating efficient real-time monitoring capabilities. This research provides an innovative solution for optoelectronic-enabled battery management systems, realizing the synergistic integration of optical fiber sensing and artificial intelligence.