论文部分内容阅读
针对传统回声状态网络(ESN)难以解决多振荡子叠加(MSO)问题,提出一种增量式模块化回声状态网络(IM-ESN).该网络储备池由多个相互独立的子储备池组成.利用矩阵的奇异值分解(SVD)构造每个子储备池的权值矩阵,并依据分块对角矩阵原理,将子储备池逐一添加至网络中.在网络增长过程中,IM-ESN无需放缩权值矩阵便能保证网络的状态回声特性.MSO问题的仿真结果表明,IM-ESN能够自主确定与问题复杂度相匹配的网络规模,具有较好的预测性能和鲁棒性.
In order to solve the problem of multi-oscillation sub-overlay (MSO) in traditional Echo State Network (ESN), an incremental modular echo state network (IM-ESN) is proposed. The network reserve pool consists of several independent sub- .Using the matrix singular value decomposition (SVD) to construct the weight matrix of each sub-reserve pool, and according to the block diagonal matrix principle, the sub-reserve pool is added to the network one by one.In the process of network growth, IM-ESN need not The shrinkage matrix can guarantee the state echo of the network.The simulation results of MSO show that IM-ESN can determine the network size that matches the complexity of the problem autonomously, and has better prediction performance and robustness.