Neural network predictive controller based on the improved TPA-LSTM model for ultra-supercritical units

Area of Science:

• * Power Systems Engineering
• * Control Theory
• * Artificial Intelligence

Background:

• * Integrating large-scale renewable energy into China's national grid necessitates enhanced flexible peaking capability in coal-fired thermal power units.
• * Existing coordinated control systems for coal-fired units struggle with multivariable coupling, slow responses, and uncertain coal quality.
• * Traditional linear predictive control methods are limited in handling disturbance uncertainties.

Purpose of the Study:

• * To develop an advanced neural network predictive controller to improve the flexible peaking capability of coal-fired power units.
• * To address challenges in multivariable coupling, response speed, and parameter uncertainty in unit load control.
• * To enhance the adaptability and robustness of control systems under varying operational conditions.

Main Methods:

• * Establishment of a data-driven control model utilizing an improved TPA-LSTM neural network.
• * Design of a multivariable coordinated control strategy leveraging the neural network controller for parameter decoupling.
• * Integration of an automatic model updating mechanism for real-time recalibration.

Main Results:

• * The proposed controller effectively handles disturbance uncertainties, surpassing traditional linear predictive control.
• * The multivariable coordinated control strategy achieved effective decoupling and high adaptability across all load conditions.
• * Simulation results confirmed the strategy's excellent control effectiveness for 1000 MW ultra-supercritical units.
• * The automatic model updating mechanism significantly improved control performance after model mismatches.

Conclusions:

• * The neural network predictive controller offers a viable solution for enhancing the flexible peaking capability of coal-fired power units.
• * The data-driven approach with automatic model updating enhances control robustness and adaptability, crucial for grid stability.
• * This strategy effectively meets the demands of integrating renewable energy sources by ensuring reliable power unit operation.