Railway wire harness connectors are critical elements in modern rail transport systems, ensuring reliable signal transmission, power distribution, and communications across the subsystems that govern traction, braking, and passenger information. The progressive deterioration of these connectors under harsh operating conditions, particularly temperature variations encountered during continuous railway operations, poses significant challenges to system reliability and operational safety. This paper presents a hybrid framework integrating an adaptive Wiener process with a deep generative model (DGM) for reliability assessment and remaining useful life (RUL) prediction of railway wire harness connectors under multi-temperature conditions. The proposed methodology combines Arrhenius-based temperature acceleration with a Wiener degradation model that characterizes temperature-dependent degradation kinetics. Specifically, a variational autoencoder (VAE) is employed as the deep generative network to learn the complex nonlinear degradation patterns that conventional parametric models may fail to capture. Furthermore, a particle filter algorithm is incorporated to enable real-time Bayesian parameter updating and state estimation, thereby allowing the model to be refined in an adaptive manner as new monitoring data become available. The effectiveness of the proposed method is validated through accelerated degradation tests on electrical connectors at four temperature levels (25°C, 55°C, 85°C, and 105°C), demonstrating that the RMSE is reduced by 23.5%, 18.2%, and 32.1% compared with the standard Wiener process, LSTM-based approach, and Gaussian process regression, respectively. The analytically derived reliability function and RUL distribution provide comprehensive uncertainty quantification to support maintenance decision-making in railway systems.