Abstract
UDC 621.3.01: 514.8: 004.8
DOI https://doi.org/10.52577/eom.2026.62.3.47
The calculation of two loads of a four-port circuit with an unstable resistance of a common wire is considered. The feedforward neural network training data represent sets of the loads, base values, and corresponding input current values. They are divided into training, validation, and test sets with some change steps for the values. In the training epochs, the neural network reveals this internal pattern in those three sets and shows small errors. However, the errors appear for the extended control data in different step types. Combining training data with different change steps eliminates this pattern. The apparent use of one base current at each input to generate the input vector results in unsatisfactory training results. In turn, the redundant two quantities of the base current at one of the inputs radically increase the accuracy and capability of generalization. The results obtained develop the methods of neural networks and provide a basis for considering multi-ports with a large number of loads.
Keywords: multi-port, load calculation, neural network, training data, relative error.