基于进化规划的自适应高斯神经网络

自适应高斯神经网络能够对目标信号的功率谱有效识别特征进行自动提取和分类,但此网络使用BP算法,其误差能量函数是一个不规则的超曲面,容易陷入局部极小值.因此,提出了一种使用进化规则来设计和训练自适应高斯神经网络的新方法.该方法能够自动地确定网络的最优结构和联结权值,同时避免网络的局部优化.实验结果表明,将该方法用于被动声纳目标的分类识别,能够有效地克服局部最小问题,具有更好的识别率.

Adaptive Gauss neural network based on evolutionary programming

The adaptive Gauss neural network can pick up recognition characteristics from the ship noise spectrum and classify the ship noise effectively, but the error energy function of back-propagation algorithm used in the adaptive Gauss neural network is an irregular hypersurface, and the network always falls in the local minimum. Thus, an adaptive Gauss neural network based on the evolutionary programming is proposed, which can optimize the structure and adjust the weight of the network automatically. This kind of adaptive Gauss neural network is used to classify the passive sonar target, and the result of experiment shows that the adaptive Gauss neural network based on the evolutionary programming can solve the local minimum problem and be more effective in classification.