Establishment of Simulation Model for Automotive LPG Engine

1.1 machine model according to the study

LPG engine technical parameters and structural measurement parameters, combined with thermodynamic models of LPG engines, use GT-power software, engine fuel supply systems, intake systems, cylinders, booster systems, exhaust systems, crankcases, and corresponding boundary conditions And LPG's fuel modeling.

1.3 Verification of the validity of the model

In order to verify the validity of the established model, the external characteristics simulation results of the model under standard atmospheric pressure were compared with the experimental data. The maximum error between the calculated value and the experimental value was not more than 3%; the output torque of the engine was 500~800Nm, and the rotation speed was 1400r. At the time of /min, the maximum output torque of the engine is 800 Nm; the output power of the engine increases with the increase of the rotation speed and its value is 30-170 kW. The simulation results are in good agreement with the experimental data of the LPG engine, indicating that the established LPG engine model is reliable and available. Rapid simulation of steady-state and dynamic characteristics of LPG engine.

3 LPG Engine Modeling Based on Neural Network

3.1 Artificial neural network is the use of physically achievable devices or existing computers to simulate certain structures and functions of neural networks in living organisms, and then used in turn for mathematical models of algorithms in engineering or other fields. It can fully approximate any complex nonlinear relationship, and has strong robustness and fault tolerance. This study uses neural networks with three or more layers, including the input layer and the middle layer (hidden layer).

And the output layer, full connection between the upper and lower layers, no connection between each layer of neurons. After a pair of learning samples are provided to the network, the activation values ​​of the neurons propagate from the input layer through the intermediate layers to the output layer, and the input responses of the network are obtained by each neuron in the output layer; then, the direction of the target output and the actual error are reduced. From the output layer, each connection weight is corrected layer by layer through each intermediate layer, and finally it returns to the structural layer (BP algorithm) of the input BP neural network model. As this error reverse propagation correction continues, the correct rate of response of the network to the input mode continues to increase.

3.2 Determination of model structure and related functions

According to neural network theory, general problems can be achieved through a single hidden layer. Since there is only one element in the input vector, the number of neurons in the input layer is one, and there is only one output vector, so the number of neurons in the output layer is one. According to the Kolmogorov theorem, the number of neurons in the middle layer of the network can be 3-8. In order to test the influence of the number of different neurons in the middle layer of the neural network on the training results, 3 to 10 middle layer neurons were taken to model the neural network.

The neuron transfer function in the middle and output layers of the network adopts the tangent function tansig, the network training uses the function trainlm, the learning algorithm of weights and thresholds adopts learngdm, and the network performance function adopts mse. 3.3 The model Matlab calculates the realization based on the established nerves. The network model structure and sample data were compiled and the Matlab calculation program was used to calculate the model. By simulating the number of different number of neurons in the middle layer, it is found that when the number of neurons in the middle layer is 6, the simulation training process is optimal and the error is minimal.

Therefore, the number of intermediate neurons is taken as six.

3.3 Model Error Analysis and Comparison with the Effect of Interpolation Model

In order to test the effect of the neural network model, the test sample input value is input into the model, and the error between the output value and the sample true value is observed. Considering that the current hybrid engine model mostly uses linear interpolation (the method uses limited known data points, many unknown data points are obtained by linear interpolation, and then the characteristic parameters of the engine under different working conditions are obtained. Although it is simple and feasible, there is a certain amount of error. In this paper, the data error obtained by the linear interpolation model is also investigated.

Based on the error analysis results, the following conclusions can be drawn:

(1) Although the training samples of BP neural network are not enough, it completes the task of predicting the test data. The maximum prediction relative error is 5.96%;

(2) The relative error root mean square (0.0321) of the BP neural network model is obviously smaller than the relative error root mean square value (0.0570) of the linear interpolation.

3.4 Generation of Simulink Modules for Neural Network of LPG Engine

In order to apply the trained network to the in-depth study of the hybrid vehicle control strategy, the network structure was transformed into a Simulink module, which was then used as reference for related research.

Enter the gensim (netbp, -1) command on the Matlab command line to generate the Simulink module for the neural network. Since the module data is normalized, the front and back ends of the module need only be mathematically processed, and the results shown in FIG. 6 can be obtained. This neural network model can be applied to the rapid modeling of hybrid car engines.

4 Conclusion

This paper establishes the GT-power simulation model of LPG engine, simulates the dynamic performance of LPG engine using GT-power simulation model, and obtains the quantitative results; it uses the dynamic data of LPG and neural network modeling method to establish The Simulink model of the LPG engine and the error analysis with the linear interpolation method prove that the model is effective and can be applied to the research of the control strategy of the gas-electric hybrid vehicle.