Stochastic gradient descent algorithm for the predictive modelling of grate combustion and boiler dynamics.

Area of Science:

• * Boiler Engineering
• * Process Control
• * Thermodynamics

Background:

• * Grate-fired boilers face challenges in pressure and water level control due to combustion lag, especially coal-fired types with low volatile matter and high char content.
• * These control issues lead to suboptimal operation and reduced boiler performance.
• * Existing control strategies struggle with the inherent dynamics of these systems.

Purpose of the Study:

• * To develop a novel predictive and dynamic simulation model for drum dynamics analysis in grate-fired boilers.
• * To integrate data-driven and thermodynamic modeling approaches for improved boiler control.
• * To minimize pressure and water level errors through multi-objective optimization.

Main Methods:

• * Development of a hybrid model combining data-driven and thermodynamic approaches for boiler dynamics.
• * Utilizing a data-driven methodology for estimating combustion, heat transfer, and circulation performance.
• * Employing the Stochastic Gradient Descent algorithm for adaptive learning and error minimization.

Main Results:

• * The integrated model demonstrates good accuracy in predicting combustion and boiler dynamics.
• * Multi-objective optimization effectively minimized pressure and water level errors.
• * The Stochastic Gradient Descent algorithm showed rapid learning and adaptation capabilities.

Conclusions:

• * The proposed predictive and dynamic simulation model offers a robust solution for grate-fired boiler control.
• * The hybrid modeling approach enhances consistency and reduces randomness in boiler operation.
• * The model shows significant potential for controlling reciprocating grate solid-fuel boilers under variable load conditions.