Deep regression adaptation networks with model-based transfer learning for dynamic load identification in the frequency domain

Frequency-domain dynamic load identification methods based on neural network (NN) models construct independently at each frequency, but are inaccurate and inefficient to train. To address these problems, a deep regression adaptation (DRAN) with model-transfer learning is proposed for identifying loads in the frequency domain. The aim take advantage of similarity uncorrelated multi-source multi-vibration response adjacent frequencies. First, DRAN model established using historical data specific frequency. Second, trained parameters transferred target as initial parameter values. Next, fine-tuned obtain Finally, current next This process iterated until all frequencies established. Because function continuous varying relationships between similar. can adapt different one training, then extract common feature information improve accuracy model. Moreover, instead setting weights randomly training them model, used better from method was evaluated experimental cylindrical shell structure under acoustic vibration joint excitation. results show that weights, higher accuracy, noise robustness, shorter time than network.