Year: 2022
Type: article
Abstract: The article is devoted to solving the issue of energy consumption planning of complex energy technological complexes. The expediency of application of self-organization methods of models of complex systems for modeling of power consumption based on the principles of inductive modeling was substantiated. Power consumption modeling was performed using neural networks based on iterative algorithms of the group method of data handling. The use of GMDH neural networks provided automatic structural-parametric identification of the model. The search for the optimal structure of the model was performed in three classes of support functions of neural networks. As a result of the algorithm of synthesis of models, the set of models-candidates having optimum structure in its class was formed. The obtained models have high modeling quality and low forecast error. Complicating the structure of the neural network does not significantly improve the quality of modeling results. The selection of the best structure of the model was made from the standpoint of multicriteria selection using a morphometric approach. Previously, all the criteria were reduced to one type according to the function of the goal and the procedure of their normalization was performed. The best structure of the model was chosen taking into account the area of the figure formed by the normalized values of the selection criteria. The polynomial GMDH neural network of the first order was chosen as the power consumption model. Deviations of actual and model values of power consumption do not exceed 1%. The usage of GMDH neural networks as an algorithm for model synthesis, supplemented by multi-criteria selection of the best model structure, allowed making an unambiguous choice of the structure of the power consumption model, suitable for determining the energy baseline.
Author: DAVYDENKO L.V.
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4290949643