Characteristics Modelling And Control Of Air Flow Rate Of Air Supply Sub-system In SOFC System Using Random Forest

Solid oxide fuel cell(SOFC)is promising to improve the environment and energy structure by reducing thermal power and compensating for the instability of new energy.Accurate and quick control of the air flow rate of air supply system is important for the efficient,economical and safe operation of SOFC systems.Response of air supply has great effects on the reaction,efficiency and life of stack.In this study,a strategy combined neural network PID with random forest model based Smith predictor is proposed to control the air flow rate of the air supply system.The main research includes the following aspects:(1)According to the design specifications and test requirements of the 5-kW solid oxide fuel cell system,a test rig of air supply system was designed and built.The performance of the test rig was verified under the open-loop condition with the host computer software developed on Lab VIEW.Performance experiments were conducted to obtain detailed performance data of the air flow rate.(2)A random forest model of the air supply system is established in the data-driven mode.Seven parameters of the inlet and outlet temperature,humidity,pressure and motor speed are selected as the input parameters of the model.The number of decision trees in the random forest is set to 592 by grid optimization.About 80%and 20%of the performance data are selected as training and test data,respectively.The results show that the model can effectively improve the prediction accuracy of air flow rate with a maximum error less than 10%.The algorithm comparison results show that the decision tree algorithm is more suitable for modelling the air flow characteristics of the air supply system.(3)A Smith predictive control strategy based on random forest and neural network PID is proposed to control the nonlinear and time-delay air supply system.The closed-loop stability of the control system is proved theoretically based on Lyapunov theorem.Then the controller was implemented using LabVIEW and Python and proved to be effective by a series of verification experiments.The experimental results show that the proposed control strategy is robust and has a faster dynamic response that shortened the settling time by 33%compare to the PI control algorithm.The overshoot and the steady error were less than 2%and 1%,respectively.