Experimental Evaluation of Horner's Method for CPU Energy Reduction in Nonlinear Modelling

Computer applications have played a central role in human progress over the past few decades. Their increasing prevalence has raised concerns about high energy consumption, drawing the attention of many scientists. Numerous studies have been conducted to develop techniques aimed at enhancing the efficiency of algorithms. Many of these studies evaluate performance based on computational time, which is measured by the elapsed time between two points in the algorithm. In this work, we present an experimental evaluation of energy reduction that reveals the computational time metric may overestimate the benefits of code efficiency. We have employed Horner's method to improve the code efficiency of two identified polynomial Nonlinear AutoRegressive Moving Average with eXogenous inputs (NARMAX) models for the Lorenz Attractor and Mackey-Glass systems. To directly measure Central Processing Unit (CPU) energy consumption, we utilised a Power Measurement Device. Our findings indicate that, while commonly used indicators such as simulation time and CPU utilisation are informative, they may not fully capture the complete picture of power usage.