Multi-task stochastic configuration network with autonomous linking and its application in wastewater treatment processeshen, the output weights of the current network

Stochastic configuration networks (SCNs) have been widely used for modeling complex industrial process due to their rapid learning speed, ease of implementation, and universal approximation capability. For modeling water quality parameters in wastewater treatment processes (WWTP), however, multiple complex tasks are often required to be modelled simultaneously. In this paper, a multi-task stochastic configuration network with autonomous linking characteristic is proposed to further develop the modeling capability of SCNs to deal with multi-tasks and achieve simultaneous measurement of multiple critical water quality parameters in the WWTP. The method can autonomously construct corresponding common nodes and proprietary nodes according to the distribution characteristics of different tasks to model the shared and private information among these tasks. Specifically, the relevant information between these tasks is explored by constructing common nodes; then personalized approximation of each task is achieved by constructing proprietary nodes for different tasks, thus improving the overall modeling performance of the model. A series of benchmark experiments and an industrial case from WWTP are carried out to verify the superiority of the proposed method. Experimental results demonstrate that our proposed method has a promising potential for multi-task data modeling.