About the article
G.A. ZUBKOV, Researcher
The article is devoted to the development of an algorithm for optimizing the hyperparameters of a multilayer perceptron for training the feedforward algorithm, which solves the problem of predicting the technical state of a spacecraft. The relevance of predicting the state of spacecraft is due to the increasing level of complexity of the tasks performed by the analysis sector of the mission control center. The proposed algorithm includes the determination of the empirical distribution of hyperparameter values with a random search for values from this distribution, as well as further search for the values of the model hyperparameters on the grid. The use of this algorithm made it possible to significantly reduce the time for finding the optimal hyperparameters in comparison with other methods of searching for hyperparameters. A qualitative assessment was made by comparing the total time of finding the hyperparameters using the algorithm and using other methods.
neural network, multilayer perceptron, hyperparameters, optimization, spacecraft, machine learning, grid search.