摘要
针对标准遗传算法优化BP神经网络收敛慢,易陷入局部最优的问题,提出了改进的多种群协同进化遗传算法,该算法改变了以往的随机初始化方法,采用了附加混沌扰动的tent映射初始化均匀分布的种群,提高了初始解的质量;每个种群采用自适应交叉率和变异率,引入移民算子实现种群间的横向联系;算法通过多种群的协同进化和种群间的个体移植提高了算法的搜索均匀性和效率;仿真实验表明该算法误差小,收敛速度快,诊断正确率高,较好地解决了模拟电路的软故障诊断问题。
The Standard Genetic Algorithm has slow convergence of BP neural network optimization, easy to fall into the local optimum problem. For this, it propose an improved multi--population co--evolutionary genetic algorithms. The algorithm changes the random initialization method. Using a tent map which addition chaotic disturbance initializes uniformly distributed population, improve the quality of initial solution initialization. Each population using adaptive crossover rate and mutation rate, introduce the immigrant operator to achieve horizon- tal linkages between population. The optimal solutions of each generation into the posts population. A variety of groups of co-- evolution and individuals transplant between populations increased the search range and efficiency. The simulation shows that the algorithm has fast convergence, High diagnosis accuracy rate and it better solutes the soft fault of analog circuit.
出处
《计算机测量与控制》
CSCD
北大核心
2012年第6期1483-1485,共3页
Computer Measurement &Control
关键词
神经网络
遗传算法
多种群协同进化
混沌
故障诊断
neural network
genetic algorithm
multi-- population co-- evolutionary
chaos
fault diagnosis