Marine power system temperature sensors play a pivotal role in engine health monitoring and energy efficiency optimization, but their accuracy is severely challenged by nonlinear thermal drift, vibration-induced noise, CANbus crosstalk, etc. Pt100 thermistors dominate over 80% of marine monitoring applications, yet traditional polynomial and piecewise linear compensation models fail to address hysteresis, parameter coupling, and sensor aging. This paper proposes a novel hybrid model integrating Differential Evolution (DE) and Support Vector Machines (SVM) for dynamic error compensation, The proposed method addresses the non - linearity and inaccuracy issues of temperature sensors by establishing a precise mapping between temperature variations and sensor outputs. Sample data was collected through experiments and simulations were conducted, the results indicate that the model enhances temperature measurement accuracy, facilitates precise fuel consumption modeling, supports real-time CO₂ emission calibration for EEOI compliance, and reduces operational costs for marine hybrid propulsion systems.