Optimizing Multi-Level Industrial Systems through Archive-Based Data-Driven Configuration

The paper presents a refined methodology for multi-criteria optimization of hierarchical production processes, leveraging Pareto-front analysis through non-dominated sorting. The proposed approach is aimed at increasing the efficiency of the system by considering mutually conflicting objective functions to guide the selection of configuration parameters towards maximizing the possibility of having future values of these objective functions lie on the Pareto front. An example of a boiler unit of a Combined Heat and Power plant (CHP) is considered, and a corresponding hierarchical boiler model is built and investigated with use of historical data. As an example of objective functions, the efficiency of the boiler unit and the consumption of overheated steam are considered. Moreover, the paper examines two definitions of the concept “hierarchy” and demonstrates their practical application, based on the data overflow through the multi-level model of the control object. Our proposed methodology applies a top-to-down non-dominated sorting algorithm through a quantized version of the data, for determining the optimal configuration parameters across different levels of a system's hierarchy. Our findings show the perspectives of updating the different setups of a system's configuration parameters according to the desired region of work, to optimize the overall possible outcome.