An integrated dynamic-evaluation and multi-task deep learning framework for load forecasting and compressor...

Area of Science:

• Industrial Engineering
• Energy Systems
• Artificial Intelligence

Background:

• Industrial compressed-air systems exhibit complex nonlinear dynamics, impacting energy efficiency and stability.
• Current methods often separate load forecasting from scheduling and lack data quality assessment, reducing reliability.

Purpose of the Study:

• To develop an integrated data-driven framework for load forecasting, dynamic evaluation, and scheduling recommendations in industrial compressed-air systems.
• To enhance energy efficiency and operational stability by addressing the limitations of decoupled forecasting and scheduling methods.

Main Methods:

• Developed a particle swarm optimization (PSO)-enhanced long short-term memory (LSTM) model for short-term load forecasting.
• Implemented a dynamic evaluation module to assess pressure compliance and fluctuation, creating tiered quality datasets.
• Constructed a multi-task fully connected neural network for optimizing compressor loading strategies based on evaluated data.

Main Results:

• The integrated framework demonstrated high forecasting accuracy and significantly improved scheduling performance.
• Utilizing the top 50% quality dataset reduced specific energy consumption by 9.26% and increased the effective operating ratio by 3.28%.
• The framework successfully integrated forecasting and data-quality control for robust scheduling.

Conclusions:

• The proposed data-driven framework offers a generalizable solution for optimizing industrial energy systems under dynamic conditions.
• Integrating load forecasting with dynamic data quality assessment and scheduling is crucial for enhancing efficiency and stability.
• The PSO-LSTM model and multi-task neural network effectively addressed the complexities of industrial compressed-air systems.