Self-Organizing Recurrent Stochastic Configuration Networks for Nonstationary Data Modeling

Recurrent stochastic configuration networks (RSCNs) are a type of randomized learner models that have shown promise in modeling nonlinear dynamics. In many fields, however, the data generated by industry systems often exhibit nonstationary characteristics, leading to the built model performing well on the training data but struggling with the newly arriving data. This article focuses on developing a self-organizing version of RSCNs, termed as SORSCNs, to improve the continuous learning capacity of the network for handling nonstationary data. SORSCNs can autonomously adjust the model parameters and reservoir structure in response to real-time data streams. The output weights are updated online using the projection algorithm, while the network structure is dynamically adjusted in the light of the recurrent stochastic configuration algorithm and an improved sensitivity analysis. Comprehensive comparisons among the echo state network (ESN), online self-learning stochastic configuration network, self-organizing modular ESN, RSCN, and SORSCN are carried out. Experimental results from two industrial predictive analyses highlight the superiority of the proposed SORSCNs in generalization performance, indicating great potential for modeling nonlinear systems with nonstationary dynamics.