Scholarlink

This research paper shows the credibility of Long Short-Term Memory (LSTM) based machine learning techniques to emphasize their use in modeling nonlinear systems. This is proposed after LSTM showed superior results when used for modeling compared to other neural networks such as ELMAN, JORDAN, and DIAGONAL. LSTM further served refined outputs when paired with db-4 (Daubechies 4) wavelets as observed for various parameters such as MSE (Mean Squared Error), MAE (Mean Absolute Error), RMSE (Root Mean Square Error), and R2 Score (The Coefficient of Determination). The observations were recorded by comparative graphs of training loss at 50 epochs and errors versus iterations of different models for 200 iterations, keeping sampling time as 0.45 s, where the graphs were mostly steep or observed closest to x-axis for LSTM. Hence, supporting the potential of LSTM-based approaches for data-driven modeling of complex dynamical systems.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep Learning for Nonlinear MISO Systems: LSTM-Based System Modeling

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Kumud Chaprana,
  • Pinaki Jha,
  • Farhaan,
  • Yash Kumar,
  • Richa Sahu,
  • Smriti Srivastava

摘要

This research paper shows the credibility of Long Short-Term Memory (LSTM) based machine learning techniques to emphasize their use in modeling nonlinear systems. This is proposed after LSTM showed superior results when used for modeling compared to other neural networks such as ELMAN, JORDAN, and DIAGONAL. LSTM further served refined outputs when paired with db-4 (Daubechies 4) wavelets as observed for various parameters such as MSE (Mean Squared Error), MAE (Mean Absolute Error), RMSE (Root Mean Square Error), and R2 Score (The Coefficient of Determination). The observations were recorded by comparative graphs of training loss at 50 epochs and errors versus iterations of different models for 200 iterations, keeping sampling time as 0.45 s, where the graphs were mostly steep or observed closest to x-axis for LSTM. Hence, supporting the potential of LSTM-based approaches for data-driven modeling of complex dynamical systems.