函数连接型神经网络在氟化物非晶态形成条件中的应用(英文)

APPLICATION OF FUNCTIONAL-LINK NET TO PATTERN RECOGNITION OF AMORPHOUS FORMATION CONDITIONS OF FLUORIDES

函数增强型神经网络 (FunctionalLinkNet,简称FLN)是一种无隐含层的新型网络 ,应用其三阶联合激励增强特性来对三元体系氟化物非晶态形成条件进行识别研究 ,识别准确率近 1 0 0 %。在对预测集的每一个输入信号添加1 0 %的噪音干扰后 ,发现识别率依然不变。直到噪音添加到1 5% ,仍然能准确判别。可见网络的容错能力是十分令人满意的。

Functional-Link Net(FLN)is a single-layer neural network,without hidden layer.We use the threeorder joint-activition of the FLN to study pattern recognition of the amorphous formation conditions of trinal fluorides.The predicted classification discrimination is 100%.After adding 10% noise to every input signal of the testing set,the discrimination retains constant.The classification discrimination is still accurate until adding 15% noise to every input signal of the testing set.The fault-tolerant ability of the FLN is very satisfactory.