SN Computer Science Computer Graphics and Computer-Aided Design

Model Order Determination: A Multi-Objective Evolutionary Neural Network Scheme

Ligeiro, Rui, Carvalho, Joao Paulo
Springer Nature · 2022-04-26 · DOI 10.1007/s42979-022-01134-9

Abstract

Knowing the relevant amount of information needed to correctly predict patterns present in a data series is an important question to address. This is known as the problem of model order determination and is not adequately solved yet. This paper proposes to determine model order based on the scheme of a topological dynamic neural network that examines the dimensionality of the non-linear function that reconstructs the process. The novelty of the approach lies in the use of a neural network optimized by an evolutionary multi-objective selection mechanism that is capable of determining model order and performing robust estimations based on joint minimization of the length of the past and of the prediction error. Since the size of the input layer of the neural network is associated with the model order, the results show that the model order can be determined by the Pareto-optimal solutions that emerge from the optimization process. The practicality of the model is demonstrated on three univariate examples extracted from a time series database.

Read at publisher →