Abstract: The remaining life of power cables in underground utility tunnels is difficult to evaluate using isolated inspection temperature records because cable ageing is jointly affected by thermal stress, humidity, load variation and local installation conditions. Based on the 2024 temperature inspection records from the south and north sections of an underground utility tunnel, this study develops a hybrid remaining-life prediction model that combines an Arrhenius thermal ageing constraint with a BP neural network. After data cleaning, 1341 valid temperature samples were obtained, with an average temperature of 28.52 °C and a maximum temperature of 41.2 °C. The proposed Fusion-BP model achieved an MAE of 1.38 years under normal operating conditions and an RMSE of 2.21 years under high-load disturbance conditions. The proposed method transforms routine temperature registration data into quantifiable life margin and hot spot risk-ranking indicators, providing technical support for condition-based maintenance.