A data-driven approach for industrial utility systems optimization under uncertainty

Abstract

Energy optimization of utility system helps to reduce the operating cost and save energy for the industrial plants. Widespread uncertainties such as device efficiency and process demand pose new challenges for this issue. A hybrid modeling framework is presented by introducing the operating data into mechanism model to adapt the changes of device efficiency and operating conditions. Mathematical models of boilers, steam turbines, and letdown valves are then developed in the framework. Based on the process historical data of a real-world plant, a Dirichlet process mixture model is used to capture the support information of uncertain parameters. Bridging data-driven robust optimization (DDRO) and utility system optimization under uncertainty, a robust mixed-integer nonlinear programming (MINLP) model is developed by utilizing the derived uncertainty set. The robust counterpart of the developed model can be reformulated as a tractable MINLP problem including conic quadratic constraints that could be solved efficiently. A real-world case study is carried out to demonstrate the effectiveness of the proposed approach in protecting against uncertainties and achieving a good trade-off between optimality and robustness of the operational decisions for industrial utility systems.