基于序列输入的神经网络模型算法及应用

Algorithm and application of sequence input-based neural network model

为提高神经网络的逼近能力,提出一种基于序列输入的神经网络模型及算法。模型隐层为序列神经元,输出层为普通神经元。输入为多维离散序列,输出为普通实值向量。先将各维离散输入序列值按序逐点加权映射,再将这些映射结果加权聚合之后映射为隐层序列神经元的输出,最后计算网络输出。采用Levenberg-Marquardt算法设计了该模型学习算法。仿真结果表明,当输入节点和序列长度比较接近时,模型的逼近能力明显优于普通神经网络。

To enhance the approximation capability of neural networks, a sequence input-based neural networks model, whose input of each dimension is a discrete sequence, is proposed. This model concludes three layers, in which the hidden layer consists of sequence neurons, and the output layer consists of common neurons. The inputs are multi-dimensional discrete sequences, and the outputs are common real value vectors. The discrete values in input sequence are in turn weighted and mapped, and then these mapping results are weighted and mapped for the output of sequence neurons in hidden layer, the networks outputs are obtained. The learning algorithm is designed by employing the Levenberg-Marquardt algo-rithm. The simulation results show that, when the number of the input nodes is relatively close to the length of the sequence, the proposed model is obviously superior to the common artificial neural networks.