【作者】 李炜；

【导师】 王雷； 王新立；

【作者基本信息】 山东大学 ， 控制工程（专业学位）， 2022， 硕士

【摘要】 传统化石能源的短缺,促使世界各国都加大对可再生能源的研究与利用。质子交换膜燃料电池(PEMFC)因其低排放、高能效、低工作温度等优良特性而备受关注,是未来重要能源之一。质子交换膜燃料电池系统的建模与控制研究,对提高电池寿命、系统效率以及动态响应能力都具有重要的意义。然而质子交换膜燃料电池系统具备高度非线性、时间滞后等特点,输出受当前系统输入和历史输出共同影响,传统的建模与控制方法都难以获得良好的效果,因此本文主要针对上述问题,开展如下工作:分析PEMFC系统结构、工作原理,搭建系统实验平台。介绍了 PEMFC的内部构造及各组件特点,分析了三种极化过电位对单电池输出电压造成的损失,分析了阴阳极进气压力和电堆工作温度等系统参数对PEMFC系统输出电压的影响。搭建了 PEMFC系统实验平台,进行实验采集数据,为后续进行实验分析、模型验证与控制仿真提供数据支撑。针对PEMFC系统存在非线性和时滞性,而建模复杂的问题,提出LBF深度学习动态模型。LBF模型是将LSTM和BPNN两种网络进行融合,其中LSTM网络用于从PEMFC系统历史输出电压中提取时间滞后信息,BPNN网络用于从当前系统输入参数中提取常规信息,而后对两种网络输出进行融合通过全连接层进一步提取信息预测输出电压。通过LBF模型与BPNN模型、LSTM模型和SVR模型进行消融实验与比较研究,LBF模型可以提供良好的预测性能,平均平方误差(MSE)最低仅为1.303,相较于其他三个模型的最优值降低了 84.32%,为后续控制研究提供模型基础。针对PEMFC系统输入耦合和存在约束条件,而控制困难的问题,设计了 PEMFC输出电压的模型预测控制策略。基于建立的LBF动态模型,提出带有耦合输入约束的滚动时域优化问题,建立关于输出电压的二次型优化目标函数,并研究了基于遗传算法的滚动优化求解算法。通过仿真分析表明了模型预测控制对PEMFC输出电压进行控制的有效性与可行性,并分析了参数对于控制性能的影响。与传统PID控制方法开展对比研究,结果表明相对PID控制,模型预测控制策略在动态调节时间和最大偏差分别减小了25%和71.57%,具有更好的控制性能。更多还原

【Abstract】 The shortage of traditional fossil energy sources has prompted countries around the world to increase research and utilization of renewable energy sources.Proton exchange membrane fuel cell(PEMFC)is one of the important energy sources of the future because of its excellent characteristics such as low emission,high energy efficiency and low operating temperature.Modeling and control studies of PEMFC systems are important to improve the cell life,system efficiency,and dynamic response capability.However,the PEMFC system is highly nonlinear and time-lagged,and the output is affected by both the current system input and the historical output,so it is difficult to obtain good results by traditional modeling and control methods.Analyze the PEMFC system structure and working principle,and build the system experiment platform.The internal structure of PEMFC and the characteristics of each component are introduced,the loss of single-cell output voltage caused by three polarization overpotentials is analyzed,and the influence of system parameters such as cathode inlet pressure and stack operating temperature on the output voltage of PEMFC system is analyzed.The experimental platform of the PEMFC system was built to conduct experimental data collection and provide data support for subsequent experimental analysis,model validation and control simulation.The LBF deep learning dynamic model is proposed to address the problem of nonlinearity and time lag in the PEMFC system and the complexity of modeling.the LBF model is a fusion of two networks,LSTM and BPNN,where the LSTM network is used to extract time lag information from the historical output voltage of the PEMFC system and the BPNN network is used to extract regular information from the current system input parameters,and then the outputs of both networks are fused to extract further information to predict the output voltage through a fully connected layer.The LBF model is compared with the BPNN,LSTM,and SVR models in ablation experiments.The LBF model provides good prediction performance with a minimum mean squared error(MSE)of 1.303,which is 84.32%lower than the optimal value of the other three models,providing a model basis for subsequent control studies.The model prediction control strategy of the PEMFC output voltage is designed for the problem of the PEMFC system input coupling and the existence of constraints,while the control is difficult.Based on the established LBF dynamic model,the rolling time-domain optimization problem with coupled input constraints is proposed,the quadratic optimization objective function about the output voltage is established,and the rolling optimization solution algorithm based on genetic algorithm is studied.Simulation analysis shows the effectiveness and feasibility of model predictive control for PEMFC output voltage control,and analyzes the influence of parameters on the control performance.A comparative study is conducted with the traditional PID control method,and the results show that the model predictive control strategy has better control performance with 25%and 71.57%reduction in dynamic regulation time and maximum deviation,respectively,compared with PID control.更多还原