Development of An Intellectual Control and Management System for Oil Production Technological Processes in Real-Time
Abstract
This paper presents the development and evaluation of a four-layer intelligent control and management system for oil production technological processes in real time, applied to cottonseed and sunflower oil refining lines. The proposed system integrates a 148-sensor IoT network (sampling frequency: 100 ms), a Digital Twin module, and three machine learning models: an LSTM neural network for deodorization temperature and energy consumption forecasting (RMSE = 1.8 °C; R² = 0.94), a Random Forest classifier for raw material quality categorization (accuracy: 94.3%), and a Reinforcement Learning agent for multi-objective optimization of energy consumption, product quality, and equipment wear. A hybrid adaptive control algorithm combining PID, ML feed-forward, and Model Predictive Control (MPC) is formulated and validated on three years of production data (n = 2,847,000 observations). Comparative testing against traditional PID-based systems demonstrated an 18–27% reduction in energy consumption, a 14.6% improvement in the product quality index, an 82.9% improvement in deodorization temperature control accuracy (from ±8.2 °C to ±1.4 °C), and a 61.7% reduction in unscheduled equipment shutdowns. The Digital Twin module enables predictive maintenance with an average fault-detection lead time of 72 hours and reduces electricity costs by 8–12% through virtual scenario planning. The system automatically identifies oil type and enforces oil-specific regulatory temperature limits in compliance with EU 2023/2229 glycidyl ester standards. Integration of heat recovery, enzymatic degumming, and RL-based energy management achieves a total reduction of 35–55% in specific energy consumption across all oil types.