东华大学学报（自然科学版）
東華大學學報（自然科學版）
동화대학학보（자연과학판）
JOURNAL OF DONGHUA UNIVERSITY(NATURAL SCIENCE)
2010年
2期
185-190,201
,共7页
喷射器%神经网络%蚁群优化算法%连续优化%性能预测
噴射器%神經網絡%蟻群優化算法%連續優化%性能預測
분사기%신경망락%의군우화산법%련속우화%성능예측
ejector%neural network%ant colony optimization algorithm%continuous optimization%pedormance prediction
以现有的喷射器实验数据集作为样本,用单隐层前向神经网络预测喷射器的性能,网络的训练分别采用连续蚁群系统(CACS)算法和连续蚁群优化(ACOR)算法.数值实验结果显示,用这两种蚁群算法所训练的神经网络对于喷射器性能的预测精度能够满足实际工程的要求,其中ACOR算法的训练误差小于一般的BP算法,预测精度也有所提高.
以現有的噴射器實驗數據集作為樣本,用單隱層前嚮神經網絡預測噴射器的性能,網絡的訓練分彆採用連續蟻群繫統(CACS)算法和連續蟻群優化(ACOR)算法.數值實驗結果顯示,用這兩種蟻群算法所訓練的神經網絡對于噴射器性能的預測精度能夠滿足實際工程的要求,其中ACOR算法的訓練誤差小于一般的BP算法,預測精度也有所提高.
이현유적분사기실험수거집작위양본,용단은층전향신경망락예측분사기적성능,망락적훈련분별채용련속의군계통(CACS)산법화련속의군우화(ACOR)산법.수치실험결과현시,용저량충의군산법소훈련적신경망락대우분사기성능적예측정도능구만족실제공정적요구,기중ACOR산법적훈련오차소우일반적BP산법,예측정도야유소제고.
A model of the feed forward artificial neural network(ANN)with single hidden layer is established to predict the performance of questioned ejectors,of which a great number of experimental data has been obtained in advance.Two different algorithms of continuous ant colony system(CACS)and ant colony optimization in Rn(ACOR)are adopted,respectively,for network training based on the obtained dataset.Numerical experiments show that the predictions of ejector performance by the ANN trained with the two algorithms can meet the requirement of accuracy in ejector practice,and the training error of ACOR algorithm is smaller than that of conventional BP training algorithm,resulting in higher prediction accuracy.
