Optimal Control of Specification in LPG Blend: A Deep Learning and PSO-Driven Framework for Minimizing Off-Spec...

Area of Science:

• Petroleum Refining
• Chemical Engineering
• Artificial Intelligence

Background:

• Accurate Liquefied Petroleum Gas (LPG) specifications are vital for its diverse applications.
• Variations in production unit operations can lead to off-spec LPG, causing significant economic losses.
• Current detection methods only identify off-spec issues after substantial product blending has occurred.

Purpose of the Study:

• To develop a proactive system for mitigating off-spec LPG production risks.
• To leverage deep learning and optimization techniques for enhanced refinery operations.
• To reduce economic losses and improve efficiency in LPG manufacturing.

Main Methods:

• Implemented a two-stage approach combining an LSTM deep learning model with particle swarm optimization.
• The LSTM model predicts product specifications in the blending tank using historical data.
• Particle swarm optimization identifies optimal controllable operational parameters in production units.

Main Results:

• The integrated model provides early warnings for potential off-spec LPG formation.
• Optimized operational parameters effectively mitigate off-spec risks in the final product.
• The methodology accounts for component-specific variable impacts, enhancing precision.

Conclusions:

• The proposed AI-driven framework significantly reduces economic losses associated with off-spec LPG.
• This approach enhances production efficiency and minimizes manual labor for quality control.
• The methodology is adaptable to similar refining processes, offering broad applicability.