期刊名称：TELKOMNIKA (Telecommunication Computing Electronics and Control)
印刷版ISSN：2302-9293
出版年度：2016
卷号：14
期号：3
页码：923-932
DOI：10.12928/telkomnika.v14i3.2756
语种：English
出版社：Universitas Ahmad Dahlan
摘要：In view of the improvement measures of the optimization control algorithm for the ignition system of the hydrogen-fueled engine, the L-M neural network algorithm, Powell neural network algorithm and the traditional BP neural network algorithm are used to optimize the ignition system. The results showed that L-M algorithm not only can accurately predict the hydrogen-fueled engine ignition timing, but also has high precision, high convergence speed, a simple model and other outstanding advantages in the training process, which can greatly reduce the workload of human engine bench tests. Only a small amount of engine bench test is carried out, and the obtained sample data can be used to predict the ignition timing under the whole working conditions. The mean square error of the optimization results based on L-M algorithm arrives at 0.0028 after 100 times of calculation, the maximum value of absolute error arrives at 0.2454, and the minimum value of absolute error arrives at 0.00426.
其他摘要：In view of the improvement measures of the optimization control algorithm for the ignition system of the hydrogen-fueled engine, the L-M neural network algorithm, Powell neural network algorithm and the traditional BP neural network algorithm are used to optimize the ignition system. The results showed that L-M algorithm not only can accurately predict the hydrogen-fueled engine ignition timing, but also has high precision, high convergence speed, a simple model and other outstanding advantages in the training process, which can greatly reduce the workload of human engine bench tests. Only a small amount of engine bench test is carried out, and the obtained sample data can be used to predict the ignition timing under the whole working conditions. The mean square error of the optimization results based on L-M algorithm arrives at 0.0028 after 100 times of calculation, the maximum value of absolute error arrives at 0.2454, and the minimum value of absolute error arrives at 0.00426.
