.png){kind=icon}
.png){kind=icon}
Return
Structure parameter estimation method for microwave device using dimension reduction network
.png){kind=icon}
PRE
AI
DOI：10.1007/s13042-022-01698-1.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Gaussian process (GP) is a multi-layer perceptron neural network (NN) with infinite units in its hidden layer that could learn effectively, so as a machine learning (ML) method, it has been a sought-after surrogate model in electromagnetics (EM) field. GP can implement complex covariance functions or insert itself into a more complicated probability structure to improve predictive function. When handling the high-dimensional data, however, GP still causes some problems, e.g., precise and efficiency degradation. Hence, we propose a dimension reduction network structure based on GP to strengthen its ability of feature extraction and apply the approach into the inverse surrogate models of microwave devices. In this model, the compressed representation process of input features is dependent on a multi-layer extreme learning machine (ML-ELM), which stacks multiple ELMs-based autoencoders (AEs) by nonorthogonal random parameters including weights and biases of hidden layers. The novelties of this paper are as follows: (1) enormous unlabeled data satisfied with designed requirements are generated randomly, and they are then used to train the parameters of ML-ELM except the last layer to improve its performance, instead of random determination of its weights and biases; (2) parameters of the last layer of the ML-ELM and hyperparameters of GP are fine-tuned by particle swarm optimization (PSO) algorithm based on some labeled data; and (3) we also study on the ML-ELM architecture deeply and draw some rewarding conclusions based on different experimental results of some typical testing functions. Empirical study on the triple-band microwave antenna demonstrates that compared with GP, predictive accuracy of the proposed method is high, considering about compressing for dimensionality of the same data.
Keywords:
Gaussian process
Multi-layer extreme learning machine
Autoencoder
Microwave device
Journal
IF:
2.7
Papers:
3.1K
Citations:
5.6K