Incorporating prior knowledge in solving system identification problem with insufficient samples based on pareto optimality concept

Abstract

Non-linear modeling based on limited samples is a difficult problem. Incorporating a prior knowledge to this type of problem might offer a promising solution. Various techniques have been proposed to incorporate prior knowledge but depend on one optimal solution which subject to pre-selection of coefficients. Incorporating the knowledge based on Pareto optimality concept offers simple post-selection of solutions. Yet, the proposed Pareto optimality concept may trap to either under-fitting or over-fitting problem based on the obtained Pareto front. The focus of this study is primarily to improve the initialization of the chromosome in order to obtain a reliable Pareto front. One system identification of control engineering problem is used as a problem to be validated. It is shown that the proposed technique is possible to be implemented by capturing the best solution in the obtained Pareto front and relatively improve the accuracy up to 8% performance of the prediction.