1.Compressor will return the air to the lumen preheated fresh air compressed by the air supply lumen into the cathode of the fuel cell to provide fuel cell with a steady flow of fuel. Air compressor PEMFC air supply system is a key component, as in the automotive PEMFC air supply system, the compressor and its drive motor are required small size, light weight, but must also meet the required power power and torque , Therefore, the choice of Switzerland imported by the brushless DC motor driven screw air compressor. Screw-type air compressors work efficiently over a wide range of flow rates and have a high compression ratio without oil lubrication and are well suited to the fuel cell's need for large amounts of oxygen. Air compressor for the fuel cell stack to provide a certain pressure, a certain flow of net air, by changing the air compressor speed, you can change the amount of air into the stack in response to changes in stack output power, so as to effectively track the electric vehicle load Variety. The performance of the screw air compressor shows that the pressure ratio (psm / patm), the air mass flow factor (mcpT1 / P1), the rotational speed factor (n / T1) and the air compressor efficiency have a certain relationship. Therefore, when the air compressor pressure ratio is a certain value, the air mass flow factor (air flow rate) flowing into the stack changes as long as the rotational speed factor (rotational speed) of the motor is controlled, and the output power of the fuel cell naturally becomes With the change, so as to meet the electric vehicles in different conditions, different road traffic demand.
2 air compressor mechanism modeling Air supply lumen pressure and air flow is directly related to the different air flow corresponding to different air pressure. To understand the PEMFC air intake and reaction volume within a certain period of time, it should be clear that the air flow changes, the conservation of mass and energy conservation law 9 available air supply lumen pressure changes From the above analysis shows that the establishment of air compressor The pressure control mathematical model is very complicated. All the parameters are time-varying, nonlinear and coupled variables. Some parameters set for analysis will affect the accuracy and real-time of the model. The following analysis of the air pressure control system suitable for air compressor experimental model.
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 Science and Technology. For the convenience of research, it is assumed that the hydrogen supply is sufficient and the air is fully humidified before entering the stack.
3.1 Neural Network Identification Model PEMFC running, air compressor air pressure and air flow, air temperature is a non-linear relationship, according to the measured data, using neural network fitting method to establish the dynamic model of air compressor.
When the PEMFC output power changes, the air compressor speed changes rapidly, the air flow quickly respond to changes in output power to meet the needs of the load, this time, the stack temperature increases, the air into and out of the stack temperature also increased . As the air pressure and air flow, air temperature is directly related to the air pressure will also change.
The input variables to the neural network identification model are the air flow and the air temperature of the air compressor that vary with the stack output power and the output variable is the air pressure of the air compressor. According to the experimental data, the air pressure of air compressor can be fitted with the curve of air flow and air temperature by neural network.
3.2 Experimental Data In this paper, 2040 sets of experimental data as a training sample, part of the test data.
It can be seen that there is a great difference between the orders of vectors in the original sample. In order to facilitate the calculation and prevent the supersaturation of some neurons, the input of the samples is normalized in the study so that the theoretical data lie in the range of<0,1>.
3.3 Simulation RBF neural network radial base expansion constants selected as 0.01.Elman neural network structure is 2-11-1, the middle layer of neurons using hyperbolic tangent S-type transfer function tansig, the output layer using the S-type output function logsig. Function learngdm, training function trainlm, performance function mse, training steps selected as 1000. Using the same set of experimental data training RBF neural network and Elman neural network, available air pressure neural network fitting curve. '*' Indicates the actual output curve of air pressure relative to air flow and air temperature, and solid line indicates the fitted curve of air pressure output by air compressor based on neural network with respect to air flow and air temperature.
RBF neural network fitting error than the Elman neural network fitting error is small, good fitting effect, the selection of RBF neural network to establish pressure control model of air compressor, and the actual air compressor model has a greater similarity, therefore Control deviation is small.
In addition, the simulation results show that the training time of RBF neural network is 6.27s, while the training time of Elman neural network is 17.85s, which shows that the training time of RBF neural network is short and meets the requirements of real-time system.
Therefore, the selection of RBF neural network to establish pressure control model of air compressor, control effect is good.
4 Conclusion
In this paper, 50kW fuel cell engine as the background, using RBF neural network and Elman neural network modeling of two neural networks, respectively, the pressure on the air compressor modeling. From the fitting error curve of neural network, we can see that using RBF neural network to establish pressure control model of air compressor has small fitting error and short training time to meet the real-time requirements of actual air pressure control system.