.png){kind=icon}
.png){kind=icon}
Return
An integrated dynamic-evaluation and multi-task deep learning framework for load forecasting and compressor scheduling in nonlinear industrial air systems
.png){kind=icon}
OA
AI
DOI：10.1038/s41598-026-61315-x.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Industrial compressed-air systems are characterized by nonlinear and dynamically coupled behavior, where load fluctuations and discrete compressor states jointly affect energy efficiency and operational stability. However, existing methods often decouple forecasting from scheduling and lack data-quality assessment, limiting their robustness. This paper proposes an integrated “load forecasting–dynamic evaluation–scheduling recommendations” data-driven framework. A particle swarm optimization (PSO)-enhanced long short-term memory (LSTM) model is developed for short-term load forecasting. A dynamic evaluation module is designed to quantify pressure compliance and fluctuation characteristics, enabling the construction of four training datasets with different quality levels (full dataset, and subsets containing the top 75%, 50%, and 25% samples ranked by the evaluation score). Based on these datasets, a multi-task fully connected neural network is constructed to optimize compressor loading strategies. Experimental results show that the proposed framework achieves high forecasting accuracy and significantly improves scheduling performance. Among the four datasets, the top 50% dataset yields the best results, reducing specific energy consumption by 9.26% and increasing the effective operating ratio by 3.28%. The proposed framework integrates forecasting and data-quality control into scheduling, providing a generalizable solution for industrial energy systems under dynamic conditions.
Journal
IF:
3.9
Papers:
27.2W
Citations:
83.5W