Research on Chemical and Pharmaceutical Process Optimization and Intelligent Control Based on Big Data and Deep Learning Models

Optimization of chemical and pharmaceutical processes and smart control are imperative in order to achieve the highest efficiency, yield, and quality with the least waste and energy consumption. Conventional techniques such as Partial Least Squares (PLS) and Principal Component Analysis (PCA) are not capable of extracting the rich, non-linear, and highdimensional patterns in the intricate process data and thus lead to poor performance and inconsistent product quality. In order to break these limitations, this paper suggests a new smart control system which maximally exploits big data analytics, going beyond traditional algorithms and making use of advanced deep learning models. In particular, Temporal Convolutional Networks (TCNs) are used to capture intricate temporal dependencies and long-distance patterns of time-series data, while Graph Neural Networks (GNNs) automatically learn spatial and relational features from connected process units. Trained and extensively calibrated on a benchmark industrial dataset, the Tennessee Eastman Process, the new system possesses important novelty and effectiveness. It exhibits a very high predictive accuracy of 98.2 % for such important quality variables as product composition, temperature, pressure, and flow and achieves a 25% reduction in operational variability.