一个新的时序神经网络模型及其仿真

A New Time Series Neural Network Model and Its Simulation

作者描述了一种新型的利用神经元的时间总和效应构成的用于处理时间序列的神经网络模型。我们考虑了三种可能的网络结构并着重研究了其中的一种，发现它在仿真实验中执行得很好，而且如果时间序列是由高维单位矢量构成的，则学习过程将变得比较简单，因此在附录中我们给出了两种将任意时间序列转变为高维单位矢量的方法。本文最后描述了证实这个网络特性的仿真实验并可看出此网络能确切地重复再现训练序列。

The authors propose a new neural network utilizing time summing effect of neuron for processing time series, Three kinds of possible architectures are considered in this paper and we stress one of them. lt. performs very well in simulation expedients. lt. has been found that the learning process will become rather simple if the time series is composed of high dimension unity vector. Two methods for translating any time series into high dimension unity vector are given in the appendix. In the last part of this paper, simulation exeriments which confirmed the features of this network are described. It. can be seen that this network can reproduce the training series exactly.