Selected Aspects of Neural-Evolutionary Modeling of Prices on the Day Ahead Market of TGE S.A.

Abstract

Modeling in the context of Artificial Intelligence (AI) is using mathematics to describe, analyze, and predict real-world systems. Building models that can simulate or predict various aspects of reality is a key issue that is the subject of many studies. The quality of models depends on many elements, starting from the architecture of the neural network itself, through the selection of teaching data in terms of the size of the sets, and the number of factors influencing the choice of the network itself. Modifications of the network training methods themselves also play an important role, e.g. through the use of Evolutionary Algorithms (AE). The paper focuses on several selected aspects related to the quality of modeling based on prices on the Day Ahead Market (DAM). The influence of network architecture factors, network type, number of training data, and Evolutionary Algorithms on the improvement of the model quality measured by the Mean Squared Error (MSE) and the coefficient of determination (R2) were considered.