Deep-Learning-Based Model Predictive Control of an Industrial-Scale Multistate Counter-Flow Paddy Drying Process.

Area of Science:

• Agricultural Engineering
• Control Systems
• Artificial Intelligence

Background:

• Industrial paddy drying often relies on manual control, leading to inconsistent grain moisture and reduced quality.
• Model Predictive Control (MPC) offers effective process control but faces computational challenges in large-scale applications.

Purpose of the Study:

• To develop a computationally efficient deep-learning-based model predictive control (DL-MPC) strategy for industrial-scale paddy drying.
• To improve the uniformity of moisture content in discharged grains and enhance overall product quality.

Main Methods:

• A DL-MPC strategy was designed using long short-term memory (LSTM) neural networks.
• The system establishes a mapping between inlet/outlet moisture content and paddy flow velocity for multistage counter-flow systems.
• Training utilized datasets from single and multistage drying systems, followed by simulations and experimental verification.

Main Results:

• The DL-MPC system demonstrated a significant improvement in computational speed compared to traditional MPC.
• Satisfactory control performance was achieved, with a mean absolute error of 0.190% d.b. for predicted moisture content.
• Predicted paddy flow velocity closely matched field data, validating the DL-MPC system's effectiveness.

Conclusions:

• The proposed DL-MPC strategy is effective for controlling multistage counter-flow paddy drying systems.
• This approach addresses the computational cost limitations of traditional MPC, enabling practical industrial application.
• DL-MPC enhances drying efficiency and product quality through improved moisture content uniformity.