摘要
描述了混沌BP网的拓扑结构与实现步骤,引入了模拟退火的控制策略,以便在训练后期有效地控制混沌运动.提出一种基于改进的德尔菲法的算法来确定输出样本集,以提高训练质量,得到了预期效果.给出了基于多个混沌BP网的案例匹配技术框架,提高了案例匹配的效率.与最邻近法相比,结果表明该技术对于类似钢铁生产动态调度这样的复杂系统,能够提高案例匹配的精确性,更适于解决案例属性权重难以确定或存在着耦合和复杂的非线性关系的问题.
The topology structure and implement step of chaotic BP neural network are described, and the control strategy of annealing strategy is introduced so as to control the chaotic movement during the later training. An algorithm based on improved Delphi approach is proposed to determine output sample set so as to improve the quality of training, and the expected result is obtained. The frame of case retrieving technique based on multi-chaotic BP neural network(MCBPNN) is presented, and the efficiency of case retrieving is improved. Compared with the result of the nearest neighbor (NN) algorithm, it can improve ease retrieving precision for complex system such as dynamic scheduling system for steel-making, more fitful to solve some problems in which case attributes weights are difficult to be determined exists coupling and non-linear relation among them. and it is or there exists coupling and non-linear relation among them.
出处
《系统工程学报》
CSCD
北大核心
2005年第4期410-418,共9页
Journal of Systems Engineering
基金
国家自然科学基金(7017105660084003)
国家863/CIMS(2002AA4120102002AA414610).
关键词
案例推理
案例匹配
动态调度
混沌
BP网
case-based reasoning
case retrieving
dynamic scheduling
chaos
BP neural network