An Online Strategy for Nonlinear Vector Autoregression

Nonlinear vector autoregression (NVAR), as an effective alternative to traditional reservoir computing (RC), has garnered significant attention in the field of time series prediction in recent years. Unlike conventional RC, NVAR does not require the use of random sampling matrices to define the underlying recurrent neural network, and it features fewer hyperparameters. However, NVAR is an offline model, making it challenging to meet the real-time monitoring requirements of industrial processes. Inspired by the Woodbury matrix identity, this paper proposes an online NVAR that can update the model using data arriving in blocks. Subsequently, this study constructs a soft sensing model based on the online NVAR to perform real-time monitoring of industrial processes. Finally, the effectiveness of the proposed method is validated using a real-world dataset from a wastewater treatment plant.