机车用多堆燃料电池系统健康状态评估及故障预测研究

The development and application of fuel cell locomotives will promote the great change of the rail transit system, accordingly safe and reliable operation of multi-stack fuel cell systems as the main power source of locomotives is important. Timely and effective state of health estimation and fault prediction for multi-stack fuel cell systems can provide a crucial guarantee for safe and reliable operation of the system. However, the propagation characteristic of degraded performance among multi-stack fuel cell systems is indistinct; meanwhile severe operating environments and operating conditions of locomotives is complex, which challenges the state of health estimation and fault prediction difficulties. In this project, the multi-stack fuel cell system in locomotives is focused using data-driven methods based on practical measured data. Through analyzing the degraded performance of the single-stack fuel cell system and multi-stack fuel cell systems, the propagation characteristic of degraded performance among multi-stack fuel cell systems effects on the evolution regularity of state of health of the system is analyzed. Consequently, the state of health estimation index is established. Then, a deep belief network based feature extraction and prediction modeling method is proposed. Finally, multi-state prediction is obtained. In order to ensure safe and reliable operation, reduce the operation and maintenance cost and prolong the useful life of the system, this project is to realize effective state of health estimation, accurate prediction of failure possibilities, timely fault early warning, which also can provide theoretical foundation and technical support for prognostics and health management of multi-stack fuel cell systems in locomotives.

燃料电池机车的研发及应用将促进轨道交通系统的重大变革，作为其主动力源的多堆燃料电池系统的安全可靠运行至关重要。及时有效的对机车用多堆燃料电池系统进行健康状态评估及故障预测可为系统的安全可靠运行提供重要保障。针对多堆燃料电池系统间性能衰退传播特性尚不明确、机车运行环境恶劣及运行工况复杂导致健康状态评估及故障预测难等问题，本项目以机车用多堆燃料电池系统为研究对象，利用实测数据，基于数据驱动的方法，针对单堆性能衰退、多堆间性能衰退传播，分析多堆燃料电池系统间性能衰退传播特性对健康状态演变规律的影响，建立健康状态评估指标；在此基础上，提出基于深度信念网络的特征提取、预测建模，实现多态预测。旨在实现有效评估、准确预测、及时预警，确保机车用燃料电池系统安全可靠运行、减少运行维护成本、延长使用寿命，也为机车用多堆燃料电池系统预测及健康管理提供理论基础和技术支撑。

氢能具有功率密度高、环境友好、重量轻和资源丰富等特点，促进了 能源领域的革新，目前已广泛应用于军事、轨道交通及热电联产等领域。燃料电池机车的研发与应用将促进轨道交通系统的重大变革，作为其主动力源的多堆燃料电池系统的安全可靠运行至关重要。由于燃料电池系统是一个具有多物理、多尺度和高度不确定性的复杂非线性系统，其老化衰退机理尚不清晰，并且其健康状态评估与故障预测较为困难。本项目通过获取机车用多堆燃料电池系统及实验用kW级电堆的海量运行数据，采用基于数据驱动的神经网络算法研究其老化与衰退机理，并提出了能精准快速的预测燃料电池系统剩余寿命的D-S ELM算法。本项目提出了基于深度信念网络的特征提取及预测建模，实现了燃料电池的多态预测。此外，本项目研究针对大功率多单片电堆进行多操作类型实验，研究了单片电池的故障情况、单片电池之间的异同性及其单片与电堆之间的相互关系。研究取得的研究成果如下：.（1）通过实验平台及机车用燃料电池获取海量数据，采用高精度传感器获取燃料电池系统进出口气体压强、温度及流量等数据，同时通过电气检测装置获取燃料电池系统电压、电流、功率及内阻等电气量数据。采用离散小波变换等信号处理方法将燃料电池系统数据进行约简与归一化处理。.（2）根据燃料电池系统输出性能及阻抗对其健康状态评估。通过燃料电池系统阻抗数据研究了各类操作条件下燃料电池的阻抗性能，并观察阻抗随操作条件的变化情况；根据电压均衡性研究观察燃料电池系统的运行状态及老化趋势；根据模糊理论对燃料电池的健康状态进行分类。.（3）针对燃料电池系统运行过程中最常见的水淹/膜干故障，概括归纳适于大尺度燃料电池系统的诊断指标；利用燃料电池电堆对水淹/膜干故障进行实验研究与故障诊断；提出了一种基于数据驱动方法的D-S ELM算法对燃料电池系统的寿命进行了精准的预测，可用于改善燃料电池系统的安全性与经济性效益。.（4）学术论文：已发表（含录用）论文 8 篇，其中SCI2篇，EI4篇，会议论文1篇.（5）专利：授权专利7项，申请发明专利5项。.（6）学生培养：培养硕士研究生4名。.（7）学术交流：邀请同行专家学术讲座2人次。