基于小世界回声状态网的时间序列预测

为了提高时间序列的预测精度, 提出了利用改进的小世界网络优化泄露积分型回声状态网(Leaky-integrator echo state network, Leaky ESN)的时间序列预测方法. 首先提出一个改进型小世界网络, 其加边概率是节点间距离的负指数函数. 然后, 利用加边概率直接表示Leaky ESN储备池两个神经节点的连接权值, 取值范围为0,1], 表征了节点间的连接程度. 利用这个新型小世界网络改进Leaky ESN的储备池神经节点的连接方式, 有目的地实现了稀疏连接, 减小了Leaky ESN储备池随机稀疏连接的盲目性, 提高了储备池的适应性.最后, 利用改进的Leaky ESN预测典型的非线性时间序列, 并利用Matlab仿真软件验证了本文提出方法的有效性. 与Leaky ESN相比, 本文提出的方法具有更高的预测精度和更短的训练时间.

Time Series Prediction with an Improved Echo State Network Using Small World Network

In order to improve the prediction accuracy of echo state network for time series, this paper uses a modified small world network to improve leaky-integrator echo state network (Leaky ESN) and then uses the improved Leaky ESN to predict time series. Firstly, we propose a modified small world network whose adding edge probability is a negative exponential function of the distance between the nodes. Then, we use adding edge probability to directly represent the connection weights of two neural nodes in the reservoir of Leaky ESN. The value of the connection weights belongs to 0,1] and represents the degree of connection between the two nodes. The modified small world network is utilized to improve the connection of reservoir neural nodes, which purposefully achieves the sparse connection to reduce the blindness of the original sparse connections and then improves the adaptive capacity of the reservoir. Finally, we use the modified Leaky ESN to predict a typical nonlinear time series. We use Matlab to verify the effectiveness of the proposed method. Compared with Leaky ESN, the method proposed in this paper has a higher prediction precision and a shorter training time.