论文部分内容阅读
对非线性系统辨识问题 ,以多层前向网络为模型框架 ,用本文提出的带自适应冷却进度表的模拟退火算法与Powell算法构成新型混合算法 ,来训练网络的权值。冷却进度表中主要参数是模拟退火算法的控制参数T的初值T0 和T的衰减函数。把整个迭代过程划分为若干阶段。在每个阶段结束时 ,依据网络模型误差自适应地修正下阶段的T0 (回火温度 )、T的衰减函数中的参数和迭代步长初值。上述混合算法具有很强的全局和局部搜索能力 ,显著提高了网络的辨识精度。应用表明了本文方案的有效性
For the problem of nonlinear system identification, the multi-layer forward network is used as the model framework, and a new hybrid algorithm is constructed by using the simulated annealing algorithm with adaptive cooling schedule and the Powell algorithm proposed in this paper to train the weights of the network. The main parameters in the cooling schedule are the initial values T0 and T of the control parameter T of the simulated annealing algorithm. The entire iterative process is divided into several stages. At the end of each phase, the parameters T0 (tempering temperature) and the initial value of the iteration step length in T attenuation function are adaptively corrected according to the network model error. The hybrid algorithm has strong global and local search capabilities, significantly improving the identification accuracy of the network. The application shows the effectiveness of the proposed scheme