Data-Driven Robust Optimization for the Operation Optimization of Industrial Power Station

Operation optimization of industrial power stations is a challenging task owing to the operational uncertainties associated with the variables and hyperdimensional system design. Thus, conducting the mathematical model based robust optimization for such large system is computationally prohibitive as well as expensive task. To this end, we have presented data-driven robust optimization framework that is built on machine learning based model (Data Information integrated Neural Network (DINN)) and nonlinear programming (NLP) technique. The robust optimization framework is applied for the operation optimization of 660 MW capacity power station. The DINN models are trained on the historical data of the industrial power station to model thermal efficiency, power and heat rate –three performance metrics of the power station. Extensive hyperparameters tuning and validation of the DINN models are carried out to select better predictive models for the three output variables. Later, multiobjective optimization problem is formulated to maximize thermal efficiency, power, and minimize the turbine heat rate by NLP technique. The estimated solution of the multi-objective function is perturbed by 10000 gaussian noise to evaluate the robustness of the estimated solution. The achieved robust solution offers optimal thermal efficiency of 42.11% and heat rate of 7642 kJ/kWh at power = 658 MW while the variance produced in the function remains less than 0.01 –the threshold can be set by the user. The proposed data-driven robust optimization framework presented in this paper can be a computational effective and cheap approach to conduct the operation optimization analytics of the industrial systems subjected to the various operational uncertainties.