相变冷却换热器动态热响应降阶预测方法研究

In response to the high computational cost associated with dynamic thermal response analysis of phase-change cooling heat exchangers (PCCHES) composed of porous skeletons with high thermal conductivity and phase-change materials under unsteady operating conditions, an efficient reduced-order prediction model integrating proper orthogonal decomposition (POD) and a feedforward neural network (FNN) was proposed. Temperature and liquid-fraction datasets were generated by numerical simulation through coupling of an enthalpy model and a porous media model, and the thermal storage performance was further analyzed. POD was employed for order reduction, and an FNN was further trained to establish the nonlinear mapping from heat flux boundary and time to modal coefficients, enabling rapid reconstruction of the physical fields. Results show that, at a heat flux density of 7 W cm^-2the maximum error between simulated and experimental temperatures was 8%. The use of composite porous graphite was found to significantly enhance the heat transfer capability of the PCCHEs. In the FNN, the coefficients of determination for the predicted modal coefficients of temperature and liquid fraction reached 0. 999 and 0.952, respectively. After reconstruction from the reduced-order model, the maximum absolute errors of temperature and liquid fraction were 0.6 C and 0.08. Compared with conventional simulation methods that take several hours, the proposed method reduced prediction time to the order of seconds while maintaining computational accuracy. The study provides a novel approach for efficient analysis and real-time prediction of PCCHES.