中文摘要：

首先构造宽带波束成形所需要的协方差矩阵,利用基于粒子群优化算法对核主成分分析方法和广义回归神经网络进行了优化。在对神经网络的输入变量进行降维处理后,生成多个复杂度低的泛回归神经网络模型。利用提出的基于聚类启发式集成算法求出波束成形时的权系数,既考虑了网络的差异性,又考虑了网络的正确性。仿真结果表明,提出的基于聚类启发式神经网络集成的波束形成算法在网络结构十分简单的情况下,仍然具有较好的性能。

英文摘要：

First, kernel principal component analysis （KPCA） method and the generalized regression neural network （GRNN） are optimized by using the particle swarm optimization （PSO） algorithm after the covariance matrix for beam forming is obtained. Second, optimized KPCA method is used to reduce the dimension of train samples in order to reduce the complexity of GRNN. Finally, considering both difference and correctness of every neural network weight coefficients for beam-forming are obtained by using the proposed neural network ensemble method based fuzzy clustering method （FCM） and Heuristic idea. The simulation results show that the proposed method has good performance under a very simple structure of the neural network.