Abstract
Convergence speed and steady-state source separation performance are crucial for enable engineering applications of blind source separation methods. The modification of the loss function of the blind source separation algorithm and optimization of the algorithm to improve its performance from the perspective of neural networks (NNs) is a novel concept. In this paper, a blind source separation method, combining the maximum likelihood estimation criterion and an NN with a bias term, is proposed. The method adds L2 regularization terms for weights and biases to the loss function to improve the steady-state performance and designs a novel optimization algorithm with a dual acceleration strategy to improve the convergence speed of the algorithm. The dual acceleration strategy of the proposed optimization algorithm smooths and speeds up the originally steep, slow gradient descent in the parameter space. Compared with competing algorithms, this strategy improves the convergence speed of the algorithm by four times and the steady-state performance index by 96%. In addition, to verify the source separation performance of the algorithm more comprehensively, the simulation data with prior knowledge and the measured data without prior knowledge are used to verify the separation performance. Both simulation results and validation results based on measured data indicate that the new algorithm not only has better convergence and steady-state performance than conventional algorithms, but it is also more suitable for engineering applications.
Highlights
Academic Editor: Xue-Bo Jin Received: 10 December 2020 Accepted: 28 January 2021 Published: 1 February 2021Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.Electrical devices are widely applied and densely placed in marine integrated power systems
Some blind separation algorithms based on second-order statistics, such as the algorithm for multiple unknown signals extraction (AMUSE) [14], and the second-order blind identification (SOBI) algorithm [15], use the non-zero time delay correlation function of the source signal to determine the separation matrix using matrix eigenvalue decomposition, which has the advantage of low computational effort and good stability for the source signal for any probability distribution
It can be concluded that the proposed algorithm is better than the competing algorithms ICA-EMK, and ICA-EBM. Both ICA-EMK and ICA-EBM algorithms are based on the maximum entropy principle, which requires the mean, variance, and higher order statistics of the source signal to be known in order to estimate the probability density function of the source signal more accurately, which is often difficult to satisfy in practice
Summary
Academic Editor: Xue-Bo Jin Received: 10 December 2020 Accepted: 28 January 2021 Published: 1 February 2021. Based on the above-mentioned advantages of NNs and considering the slow convergence and low separation performance of conventional BSS methods in engineering applications [31], a BSS method that combines the maximum likelihood estimation (MLE) criterion and an NN with a bias term is proposed in this study. This algorithm uses the MLE criterion to achieve unsupervised learning in a feedforward NN with a bias term, and its learning criterion adopts an improved RMSprop optimization algorithm (see Section 2.3 for details), which can quickly and accurately converge to the global minimum by changing the MLE criterion to the negative log likelihood (NLL) loss (see Section 2.2), making it possible to estimate the mixing matrix A for BSS. Because the computation of the loss function is relatively independent of the BP algorithm, various optimization algorithms can be applied to this structure to improve the algorithm performance
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
