A Data-Driven Framework for Dynamic Modeling and Predictive Control of Industrial Steam Boilers Using Neural Networks and LSTM Forecasting

The efficient operation of industrial steam boilers in a combined heat and power plant is essential for improving efficiency and reducing operating costs in energy-intensive industrial processes. Industrial steam boilers play a central role in supplying steam energy to various industrial processes. The efficiency of steam boiler operation directly influences the fuel consumption, production cost and environmental performance. Industrial steam boilers exhibit multivariable, time-varying, and nonlinear behavior, which makes it difficult to develop an accurate first-principles model. Accordingly, this research develops an integrated data-driven MPC-MCO framework for multi-criteria optimization (MCO) of steam boilers. The ANN model is used to capture the steam boiler’s nonlinear characteristics. Subsequently, the LSTM model is employed to forecast future steam demand using historical operational data. The forecasts are integrated into the MPC framework to determine control actions for the steam boiler. A multi-criteria optimization layer is then added to identify Pareto optimal operating points by considering boiler efficiency and steam productivity. The proposed framework improves boiler efficiency by 3% and reduces fuel consumption by 4%. The proposed methodology, therefore, integrates the capabilities of neural network modeling, forecasting, predictive control and Pareto optimization to improve the efficient operation of steam boilers.