Evaluierung der Leistungsfähigkeit von LSTM-Modellen für die Approximation physikalischer Systeme

This study investigates the application of Long Short-TermMemory (LSTM) models to approximate complex physical processes. LSTM models, which are characterized by their ability to processlarge data sets and detect hidden patterns, are evaluated for theirprediction capabilities. The research focuses on the advantages ofLSTM models, such as accurate predictions, as well as challengessuch as data dependency and the risk of overfitting. Experiments ona heating system show that LSTM models with optimized inputs andhyperparameters can be effectively used for predictive maintenanceand improving operational efficiency in industrial applications.Bayesian optimization is used for hyperparameter optimization, whichenables an improvement in model accuracy and a reduction inevaluation time. The results of this optimization and the knowledgegained from it offer valuable insights for the use of LSTM models intechnical and industrial areas.