基于LSTM-MPC的PEMFC运行状态建模与容错控制
Operating States Modeling and Fault-tolerant Control of PEMFC Based on LSTM-MPC
袁铁江 1郭泽林 1胡辰康1
作者信息
- 1. 大连理工大学电气工程学院,辽宁省 大连市 116024
- 折叠
摘要
质子交换膜燃料电池(proton exchange membrane fuel cell,PEMFC)具有多物理场耦合特性易产生不同故障且难以控制.为了能在故障状态下快速有效控制,提出基于模型预测控制(model predictive control,MPC)的容错控制方案.首先,以长短时记忆神经网络(long short-term memory,LSTM)的预测误差为遗传算法的适应度函数,寻优获取LSTM的最优超参数组合,基于数据驱动构建PEMFC系统在 4 种不同运行状态下的LSTM预测模型作为预测模型模块.然后,建立基于神经网络的控制器作为优化控制器模块,根据上述模块制定以PEMFC系统阴阳极输入气体压强为控制量、电堆电压为输出量的容错控制方案.最后,仿真验证LSTM预测模型与容错控制方案得到,LSTM预测模型在训练集和测试集的评估指标均方根误差(root mean square error,RMSE)指标值分别为0.0 489和0.0 558,具有较好的拟合效果.在不同故障状态下,MPC相较于传统PID容错控制方案电压恢复时间缩短 50%及以上,并在氢气泄露故障状态下,最大压降降低22.2%,证明了所提控制策略的有效性和正确性.
Abstract
Proton exchange membrane fuel cell(PEMFC)has multi-physics coupling characteristics,so PEMFC is prone to different faults and difficult to control.In order to control the fault quickly and effectively,a fault tolerant control scheme for PEMFC based on model predictive control(MPC)is proposed.First,the LSTM model prediction error is taken as the fitness function of genetic algorithm to obtain the optimal long short-term memory(LSTM)hyperparameter combination.Based on data drive,the LSTM prediction model of PEMFC system under four different operating states is established,which is used as the prediction model module.Moreover,the optimal controller based on neural network is established as the controller module,and the fault tolerant control scheme is developed,which takes the input gas pressure of anode and cathode of PEMFC system as the control quantity and the reactor voltage as the output quantity.Finally,the simulation verifies the LSTM prediction model and the fault tolerant control scheme,and the root mean square error(RMSE)index values of the LSTM prediction model in the training set and the test set are 0.0489 and 0.0558,respectively,showing a good fitting effect.Compared with the traditional PID fault-tolerant control scheme,the voltage recovery time of MPC is shortened by 50%or more under different fault states,and the maximum voltage drop is reduced by 22.2%under the hydrogen leakage fault state,which proves the effectiveness and correctness of the proposed control strategy.
关键词
质子交换膜燃料电池/数据驱动/神经网络/模型预测控制/容错控制Key words
proton exchange membrane fuel cell/data-driven/neural network/model predictive control/fault-tolerant control引用本文复制引用
基金项目
国家电网科技项目(5419-202257456A-2-0-ZN)
出版年
2024
国家科技期刊平台
NSTL
万方数据