基于GA-BP神经网络的超临界CO2传热特性预测研究

超临界二氧化碳（S-CO₂）动力循环在能源利用领域中拥有广阔的应用前景，其中超临界CO₂的传热特性对其能量转换效率至关重要。开展了超临界CO₂在水平小圆管内对流传热实验研究，并通过建立遗传算法优化的BP神经网络模型（GA-BP），对其在不同工况下的传热特性进行预测分析。实验参数范围：系统压力7.5~9.5 MPa，质量流速1100~2100 kg/（m²?s），热通量120~560 kW/m²。实验结果表明，超临界CO₂传热系数随流体温度的升高先增大后减小，在拟临界温度附近达到最大值。GA-BP神经网络模型能有效地预测超临界CO₂的传热系数，预测数据的决定系数R²为0.99662，超过95%的数据误差位于±10%范围内，平均误差为3.55%，为超临界流体传热预测提供新的思路。

Prediction of heat transfer characteristics for supercritical CO2 based on GA-BP neural network

Supercritical carbon dioxide (S-CO₂) power cycle has a broad application prospect in the field of energy utilization, in which the heat transfer of supercritical CO₂ plays a significant role in its energy conversion efficiency. Therefore, an experiment has been conducted to determine the convective heat transfer characteristics of supercritical CO₂ flowing in a mini horizontal circular tube, and a BP neural network optimized by genetic algorithm has been established to predict the heat transfer coefficient of supercritical CO₂ under different conditions. The experimental parameters are as follows: system pressure 7.5—9.5 MPa, mass flux 1100—2100 kg/(m²?s), heat flux 120—560 kW/m². The experimental results show that the heat transfer coefficient of supercritical CO₂ increases first and then decreases with increasing fluid temperature, and reaches maximum near the pseudo-critical temperature. The model of the GA-BP neural network can effectively predict the heat transfer coefficient of supercritical CO₂, the determinate coefficient of predicted data R² is 0.99662, and more than 95% of the data are within the error range of ±10%, the average error is 3.55%. GA-BP neural network model provides a novel idea for heat transfer prediction for supercritical fluids.