@article{4608,
  author = {Xiaojuan Chen},
  title = {LSTM-based Multi-channel Convolutional Autoencoder Model for Signal Optimization and Control Strategy},
  journal = {Digital Signal Processing and Artificial Intelligence for Automatic Learning},
  year = {2025},
  volume = {4},
  number = {4},
  doi = {https://doi.org/10.6025/dspaial/2025/4/4/147-154},
  url = {https://www.dline.info/dspai/fulltext/v4n4/dspaiv4n4_1.pdf},
  abstract = {The paper proposes a deep learning based approach to enhance the efficiency, stability, and control of hybrid
renewable energy systems. It addresses challenges such as the intermittency and variability of sources like
solar, wind, and geothermal energy, which hinder optimal energy utilization and grid stability. The study
introduces an LSTM based multi channel convolutional autoencoder model that effectively extracts temporal
and spectral features from system signals, improving prediction accuracy by 9.8% over traditional CNN
models. The architecture combines convolutional neural networks (CNNs) for noise reduction and feature
extraction with long short term memory (LSTM) networks for capturing temporal dependencies. Experimental
results using real photovoltaic data from Shaoxing, China (2014–2018) demonstrate that the Conv LSTM
model achieves superior performance across key metrics MAAPE, RMSE, and MAE especially for forecasting
beyond 15 minutes. The research also highlights the importance of integrating environmental variables, such
as temperature, which shows a correlation coefficient of 0.5539 with solar output. By leveraging advanced
deep learning techniques, the system enables adaptive control strategies, demand forecasting, and optimized
energy scheduling, ultimately improving economic viability and operational reliability. Despite its promise,
the approach faces challenges in data collection, model interpretability, and real time deployment, pointing
to directions for future research.},
}