自适应神经-模糊推理系统的混合协同微粒群算法进化设计

Evolutionary Design of Adaptive Neuro-Fuzzy Inference System Based on Hybrid Cooperative Particle Swarm Optimization

在分析自适应神经-模糊推理系统(ANFIS)结构和参数特点的基础上,定义一个布尔向量L作为网络的结构参数,与原来ANFIS的前件参数集一起构成了新的前件参数集{c,σ,L},并给出了一个新的网络输入输出关系表达式.针对该输入输出表达式,提出一种用于优化ANFIS前件参数集的混合协同微粒群算法.该将参数集L和{c,σ}分别放在两个子微粒群并根据各自不同的特点应用二进制PSO和GCPSO算法进化,两个子微粒群之间的协同由定义的一个协同函数实现,而网络的结论参数依旧用最小二乘法进行优化.应用该算法进行ANFIS网络结构和隶属函数参数的自适应设计,在Henon映射产生的混沌时间序列预测中显示了良好的性能.

A new algorithm for design of adaptive neuro-fuzzy inference system （ANFIS） is proposed in this paper. In the proposed algorithm, a boolean variable L is defined as the network structural parameter and combined with the original reasoning parameter set. Under the new parameter set { c, σ, L } , a new expression relating input and output of ANFIS is obtained. Based on the new input-output expression, a hybrid cooperative particle swarm optimization （HCPSO） is proposed to optimize the parameters. In the proposed HCPSO, the parameter sets L and { c, σ } are placed in two subswarms and evolve according to binary PSO （BPSO） and guaranteed convergence PSO （GCPSO）, respectively. A cooperative function is defined to cooperate the evolvement of the two subswarms. The output parameters of the system are optimized using Least Square Error （LSE） algorithm, The proposed algorithm can be used in adaptive design of ANFIS network architecture and membership function parameters. Simulation results indicate that the algorithm shows good performance in forecasting chaotic time seouence uenerated from Henon map.