摘要
提出了一种神经网络与粒子群算法相结合的锡磷青铜水平连铸工艺参数优化方法。以水平连铸中7个主要工艺参数为优化对象,带坯成材率为优化目标,进行正交试验并以试验数据作为样本,利用神经网络建立优化参数与优化目标的非线性映射模型。利用粒子群算法对建立的模型进行优化,获得最优铸造工艺参数。选用RBF(径向基函数)神经网络,网络学习采用减聚类算法和最小二乘法,采用惯性权重动态改变策略对粒子群算法进行改进。实际生产证明,经优化的铸造工艺参数使带坯的成材率从56%提高到71%。
A method combined ANN (artificial neural network) with PSO (particle swarm optimization algorithm) to optimize the processing parameters of QSn6.5-0.1 horizontal continuous casting has been put forward taking seven main casting technological factors as optimizing parameters, and ratio of accepted products as optimizing objective. Through the orthogonal experiment of the technical factors in the casting process, the mapping model of ANN was established based on the data from above experiment. Through optimizing the model by PSO, the optimized processing parameters have been given. A learning algorithm of subtractive clustering method (SCM) was used for radial basis function (RBF) network to get parameters RBF with the help of the least-square method (LSM), and a tactics which alters the weight dynamically according to particle position and objective function values is applied to modify the PSO. Practical production verifies that it increases the ratio of accepted products from 56% to 71%.
出处
《特种铸造及有色合金》
CAS
CSCD
北大核心
2007年第4期286-288,共3页
Special Casting & Nonferrous Alloys
