摘要
数据预测作为城市计算的一部分,在帮助理解各种城市现象本质及预测城市未来中有着举足轻重的作用。回声状态神经网络作为一种新型的循环神经网络模型,广泛应用于数据预测领域。传统的回声状态神经网络由输入层、储备池和输出层3个部分组成,其储备池中具有大量稀疏连接的神经元,对输入数据进行非线性变换可输出高维的内部状态。针对高维变换在求解输出权值矩阵时的耗时问题,提出一种基于压缩感知方法的回声状态神经网络,利用测量矩阵,将高维的内部状态压缩成低维后再求解输出权值矩阵。混沌时间序列预测实验结果表明,相对于传统模型,该方法能在误差损失允许范围内,将计算时间最大程度降低到40%左右。
As a part of urban computing,data prediction plays an important role in understanding the nature of various urban phenomena and predicting the future of cities.As a new type of circular neural network model,echo state network(ESN)has been widely used in the field of data prediction in recent years.The traditional ESN consists of three parts:input layer,reservoir and output layer.The reservoir has a large number of sparsely connected neurons and outputs high-dimensional internal states through nonlinear transformation of input data.It is time-consuming to solve the output weight matrix by high-dimensional transformation.To solve this problem,this paper proposes an ESN based on compressed sensing method,which compresses the high-dimensional internal state into the low-dimensional by using a discrete cosine transform matrix,and then solves the output weight matrix.The experimental results of chaotic time series prediction show that compared with the traditional model,this method can reduce the computation time to about 35%in the allowable range of error loss.
作者
李莉
於志勇
黄昉菀
LI Li;YU Zhi-yong;HUANG Fang-wan(College of Mathematics and Computer Science,Fuzhou University;Fujian Provincial Key Laboratory of Networking Computing and Intelligent Information Processing,Fuzhou University,Fuzhou 350116,China)
出处
《软件导刊》
2020年第4期9-13,共5页
Software Guide
基金
国家自然科学基金项目(61772136)。
