원문정보
초록
영어
Elman neural network was utilized to accomplish mapping from vibrant acceleration space of unsprung mass to dynamic load space in order to identifying results of neural network can more approach dynamic changing course of air suspension during active controlling process of air suspension. The dynamic model of 1/4 engineering vehicle was established. Vibrant acceleration data and dynamic load data were got by simulation based on the dynamic model of 1/4 engineering vehicle. Vibrant acceleration data were selected to be input data and dynamic load data were selected to be output of Elman neural network. Elman neural network was trained by input and output data. Then, generalization of trained Elman neural network was tested as follows. Sine wave was selected as road input. When amplitude was selected as 0.1m and frequency was selected as 1 rad/s, data of identifying error rate within 30% took 75.97% in total data. When amplitude was selected as 0.05m and frequency was selected as 0.5 rad/s, data of identifying error rate within 30% took 96.1% in total data. Results indicated that Elman neural network possess better fitting ability on this situation. When amplitude was selected as 0.3m and frequency was selected as 2 rad/s, results indicated that identifying error rate decreased and identifying curve obviously separated with numerical curve. It is perhaps for the reason that more identifying data scale out the value boundary of trained data. Meanwhile identifying curve and numerical curve gradually approached and stabilized eventually. It demonstrated that Elman neural network can effectively approach dynamic changing course of air suspension.
목차
1. Introduction
2. Working Principle of Air Spring
3. Air Suspension Model of 1/4 Vehicle
4. Identification based on Elman Neural Network
5. Identifying Dynamic Load
6. Conclusions
References