A PLS-based pruning algorithm for simplified long–short term memory neural network in time series prediction

作者：

Highlights：

• A hybrid simplification strategy is designed to simplify the internal structure of LSTM.

• A pruning algorithm based on PLS regression is proposed for SLSTM.

• The hidden layer size is reduced by pruning the redundant memory blocks in PSLSTM.

• PSLSTM can reduce computational complexity without performance degeneration.

摘要

•A hybrid simplification strategy is designed to simplify the internal structure of LSTM.•A pruning algorithm based on PLS regression is proposed for SLSTM.•The hidden layer size is reduced by pruning the redundant memory blocks in PSLSTM.•PSLSTM can reduce computational complexity without performance degeneration.

论文关键词：Long–short term memory (LSTM),Time series prediction,Internal structure simplification,Partial least squares (PLS) regression,Pruning algorithm,Hidden layer size

论文评审过程：Received 17 December 2021, Revised 3 August 2022, Accepted 3 August 2022, Available online 8 August 2022, Version of Record 22 August 2022.

论文官网地址：https://doi.org/10.1016/j.knosys.2022.109608