用人工神经网络模型预测钢的奥氏体形成温度  

Artificial Neural Network Applied to Austenite Formation Temperature Prediction

作　　者：由伟[1] 白秉哲[1] 方鸿生[1] 

机构地区：[1]清华大学材料科学与工程系,北京100084

出　　处：《金属热处理》2004年第10期43-47,共5页Heat Treatment of Metals

基　　金：国家科技部 863项目 ( 2 0 0 3AA3 3 10 40 )

摘　　要：根据所收集的试验数据 ,建立了预测钢的奥氏体形成温度Ac3和Ac1点的反向传播人工神经网络模型。用散点图和均方误差、相对均方误差和拟合分值 3种统计学指标评价模型的预测性能。人工神经网络预测Ac3和Ac1的 3种统计学指标分别为 2 3 8℃ ,14 6℃ ;2 89% ,2 0 6 %和 1 892 1,1 70 11。散点图和统计学指标均表明人工神经网络的预测性能优于Andrews公式。此外 ,用人工神经网络分析了C和Mn的含量对Ac3和Ac1温度的定量影响 ,计算结果表明 。The back-propagation artificial neural network was established using data collected from domestic and foreign literatures to predict the austenite formation temperatures (Ac_3 and Ac_1) of steels.Scatter diagrams and three statistical criteria--mean squared error,mean squared relative error and score of fitness were used to evaluate the prediction performance.The three criteria for predicting Ac_3 and Ac_1 using the neural network are 23.8 ℃,14.6 ℃; 2.89%,2.06% and 1.8921,1.7011 respectively.Scatter diagrams and the statistical criteria show that the prediction performance of artificial neural network is superior to that of Andrews formulae.Moreover,the quantitative effects of C and Mn contents on Ac_3 and Ac_1 temperatures were analyzed using neural network models,the results show that there exists nonlinear relationship between contents of C and Mn and the Ac_3 and Ac_1 temperatures which is mainly resulted from the interaction among the alloying elements in steels.

关 键 词：奥氏体形成温度 人工神经网络 预测性能 合金元素 定量影响 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TG14[自动化与计算机技术—控制科学与工程]