From the performance of the screw air compressor, the pressure ratio (psm/patm), air mass flow factor (mcpT1/P1), rotational speed factor (n/T1) and air compressor efficiency () have a certain relationship. Therefore, when the pressure ratio of the air compressor is a certain value, as long as the rotational speed factor (rotation speed) of the motor is controlled, the air mass flow factor (air flow rate) flowing into the stack changes, and the output power of the fuel cell is naturally It changes accordingly, so as to meet the driving needs of electric vehicles in different working conditions and different road surfaces.
2 Air compressor mechanism modeling Air supply lumen pressure is directly related to air flow, that is, different air flows correspond to different air pressures. To understand the amount of air in the PEMFC and the amount of reaction in a certain period of time, it should be clear that the air flow changes, the mass conservation and energy conservation law 9 can get the air supply lumen pressure change. From the above analysis, the air compressor is established. The mathematical model of pressure control is very complicated. Each parameter is time-varying, nonlinear, coupled variable, and some parameters that are easy to analyze, will affect the accuracy and real-time of the model. The air compressor experimental model for the air pressure control system is analyzed below.
3 air compressor experimental modeling
The fuel cell used in the experiment is a proton exchange membrane fuel cell independently developed by Wuhan University of Technology. For the convenience of research, it is assumed that the hydrogen supply is sufficient and the air is sufficiently humidified before entering the stack.
3.1 Neural network identification model When the PEMFC is running, the air pressure and air flow of the air compressor and the air temperature are nonlinear. According to the measured experimental data, the dynamic model of the air compressor is established by the neural network fitting method.
When the output power of the PEMFC changes, the air compressor speed changes rapidly, and the air flow responds quickly to the change of the output power to meet the load demand. At this time, the operating temperature of the battery stack rises, and the temperature of the air inlet and the discharge stack also rises. . Since the air pressure is directly related to the air flow rate, the air pressure also changes.
The input variable of the neural network identification model is the air flow and air temperature of the air compressor as the output power of the stack changes, and the output variable is the air pressure of the air compressor. According to the experimental data, the neural network can be used to fit the air pressure of the air compressor with the air flow rate and the air temperature.
3.2 Experimental data This paper uses 2040 experimental data as training samples and partial test data.
It can be seen that the magnitude of each vector in the original sample varies greatly. For the convenience of calculation and to prevent some neurons from reaching the supersaturation state, the input of the sample is normalized in the study, so that the theoretical data is located in the <0, 1> interval.
3.3 Simulation RBF neural network Radial basis expansion constant spread is selected as 0.01. Elman neural network structure is 2-11-1, middle layer neurons adopt hyperbolic tangent S-type transfer function tansig, output layer adopts S-type output function logsig. The function uses leanngdm, the training function uses trainlm, the performance function uses mse, and the number of training steps is 1000. Using the same set of experimental data to train the RBF neural network and the Elman neural network respectively, the neural network fitting curve of air pressure can be obtained. The middle "*" indicates the actual output curve of the air pressure with respect to the air flow rate and the air temperature, and the solid line indicates the fitting curve of the air pressure of the air compressor output based on the neural network with respect to the air flow rate and the air temperature.
The fitting error of RBF neural network is smaller than that of Elman neural network, and the fitting effect is good. The RBF neural network is used to establish the pressure control model of air compressor, which has great similarity with the actual air compressor model. The control deviation is small.
In addition, the simulation process shows that the training time of RBF neural network is 6.27s, while the training time of Elman neural network is 17.85s. It can be seen that the training time of RBF neural network is short, which meets the requirements of real-time system.
Therefore, the RBF neural network is used to establish the pressure control model of the air compressor, and the control effect is good.
4 Conclusion
In this paper, based on the 50kW fuel cell engine, two neural network modeling methods, RBF neural network and Elman neural network, are used to model the pressure of the air compressor. The neural network fitting error curve shows that the pressure control model of the air compressor using RBF neural network has small fitting error and short training time, which meets the real-time requirements of the actual air pressure control system.
DGSF SHOE MACHINERY GROUP can provide the whole leather belt making machines for leather belt making factory , and we are developing new machines continued , with 25 years making experience for Leather Belt Cutting Machine , leather strap cutter, Leather Sewing Machine , ect.
Leather Belt Making Machine
Leather Belt Making Machine,Leather Belt Cutting Machine,Leather Sewing Machine, Leather Stitching Machine
DGSF SHOE MACHINERY GROUP ,