Selection of the Learning Samples of Neural Networks for Internal-Combustion Engine Emission

According to the variation characteristics of the operating condition of internal- combustion engines, an improved orthogonal design method which considers the changes on factor boundary is proposed to choose learning samples of neural networks. The results show that the forecasting precision of the emission models built on 6135ZG diesel engine under stable operation increaseswith the increase of the level of the diagonal intersection table. Even if the sample is a diagonal intersection of level 3, the emission-forecasting model can be built with an error less than 5.7%. Compared with random selection of samples, the improved orthogonal design method is characterized by easiness and less work load.