多稀疏回声状态网络预测模型

Prediction model with multiple sparse echo state network

中文关键词  回声状态网络  稀疏  预测模型  相关向量机

英文关键词  echo state network  sparse  prediction model  relevance vector machine

中文摘要

      针对单回声状态网络难以充分描述数据信息的问题, 提出多稀疏回声状态网络预测模型. 通过对相关回声 状态网络的组合权值及由相关样本得到的基函数的权值同时进行学习, 获得优化的多个稀疏回声状态网络组合模 型. 所提模型不同于双稀疏相关向量机等多核学习模型, 它不需要选择特定的核函数及相应的核参数. 因此, 该模 型不但能更好的描述数据信息, 避免了双稀疏相关向量机及其他多核学习中核函数及其参数不易选择的问题. 同 时, 所提模型不需要采用交叉验证的方式确定回声状态网络的谱半径和稀疏度, 只需确定相应的区间. 本文通过两 组标杆数据和一组实际数据仿真实验, 与传统回声状态网络方法相比, 验证了所提模型具有更好的预测性能.

英文摘要

      Considering the problem that using a single echo state network (ESN) is difficult to describe the data information adequately, we propose a multiple sparse echo state network prediction model. The optimized combination model of echo state network is achieved by learning the sparse weights of the related ESN and the sparse weights of related basis functions determined by related sample simultaneously. And the proposed model is achieved with no need of determining the kernel functions and the related kernel parameters, which is different from the double sparse relevance vector machine and the other multiple kernel learning models. So the proposed model not only can describe the information of the datasets better but also can avoid the selection procedure of kernel functions and kernel parameters. There is no need of selecting the spectral radius and sparsity of ESN by cross validation in the proposed model and only the interval of spectral radius and sparsity are needed to be determined. The experimental results of two groups of benchmarking data and a group of real-world dataset demonstrate that the proposed model has better prediction performance.