针对SOFC/GT混合动力系统在变工况运行过程中极易出现的电堆超温、重整器碳沉积等故障，提出使用BP神经网络对SOFC/GT混合动力系统故障诊断与性能预测，基于此搭建了混合动力系统动态模型，进行了动态特性分析及验证，设计并优化BP神经网络结构，得到了超温、碳沉积故障的诊断结果和对电堆温度、系统输出功率的性能预测结果。结果表明：所设计的混合动力系统输出功率为388.4kW，效率为61.8%，满足系统设计要求。使用BP神经网络的对超温故障的诊断准确率为95%，对碳沉积故障的诊断准确率为97.5%。在空气流量阶跃降低12.5%和17.5%的工况下，对电堆温度及电堆输出功率的动态预测准确率分别达到了97.3%和98.8%。所得结果可以为未来发展长寿命的绿色高效发电技术提供技术支撑。

To address common faults such as stack overheating and reformer carbon deposition in SOFC/GT hybrid power systems during variable-condition operation, this paper proposes a fault diagnosis and performance prediction method based on a BP neural network. A dynamic model of the hybrid system was developed, and dynamic characteristics were analyzed and validated. The structure of the BP neural network was designed and optimized to diagnose overheating and carbon deposition faults and predict performance parameters including stack temperature and system output power. The results show that the designed hybrid system achieves an output power of 388.4 kW with an efficiency of 61.8%, meeting the design requirements. The diagnosis accuracy of the BP neural network reached 95% for overheating faults and 97.5% for carbon deposition faults. Under conditions of step reductions in air flow by 12.5% and 17.5%, the dynamic prediction accuracy for stack temperature and output power reached 97.3% and 98.8%, respectively. These findings provide important technical support for the development of long-life, green, and high-efficiency power generation technologies.

国家自然科学基金项目（面上项目，重点项目，重大项目）