一种基于直推置信的遗传优化概率神经网络

An optimized probabilistic neural network based on transductive confidence

提出一种基于直推置信机制的遗传优化概率神经网络用于水下航行器的机械噪声源识别.该网络模型采用遗传算法优化概率神经网络的结构,并且确定最优控制参数,在保证分类质量的同时提高了分类器的识别速度.分类器在输出端引入直推置信机制,突破了传统分类器不能有效拒识突变样本的局限,通过置信机制实现无突变噪声训练样本情况下的突变噪声的识别.实验结果表明,该网络模型具有良好的泛化性能、识别效果好,并且能有效识别突变噪声样本,是一种实用的水下航行器机械噪声源识别模型.

An optimized probabilistic neural network for underwater vehicle mechanical noise source identification is proposed in this paper.The classifier used genetic algorithm optimizing neural network structure and optimal control parameters.It improved the classifier recognition and ensured the quality of classification.The classifier can reject abnormity sample by confidence mechanism and realize training without noise samples.The experimental results show the network model has good generalization and recognition performance.And it is a practical underwater vehicle mechanical noise source identification model.