最优模糊神经网络参数的设计——混沌模拟退火学习法

Optimization design for parameters of FNN:Learning algorithm of chaos simulated annealing

提出了一种新型优化算法———混沌模拟退火学习法,将混沌算法和模拟退火算法相结合学习模糊神经网络的结构和参数。首先将混沌变量引入模糊神经网络参数的优化搜索中,利用混沌变量的遍历性寻优,根据性能指标寻找较优的模糊神经网络控制器,然后在混沌优化确定的网络基础上,把经混沌搜索后得到的全局次优解作为模拟退火学习算法的初始值,再用模拟退火方法进一步学习网络的隶属函数和权值参数,找到一个全局最优的网络。仿真结果表明:混沌模拟退火学习法优于传统优化方法,其控制结果具有精度高、超调小和响应快的优点,为解决模糊神经网络控制器参数全局最优设计提供了一种切实有效的方法。

Differently from conventional algorithms, a new optimization algorithm--hybrid optimization algorithm is proposed, which combines the chaos algorithm with simulated annealing algorithm to study parameters of fuzzy neural networks (FNN). At first, the chaos variables are applied to optimize parameters of FNN. The searching process is done by using the features of ergodicity of chaotic motion, because the chaos has no repetitiveness and randomness, and a optimal FNN is found according to the performance standard. And then, the parameters of membership function and weight is studied further by using the simulated annealing based on the chaotic optimal networks, the original values of simulated annealing are local optimal values of chaos search. Lastly, the global optimal FNN is found. It is shown from the simulation results that the hybrid optimization algorithm is superior to conventional optimization algorithm and the control results have advantages of high precision, small overshoot and fast response.