Load Frequency Control Strategy for Nonlinear Power Systems Based on DL-NMPC

High renewable energy penetration compels thermal power units to operate across wide load ranges, causing significant parameter variations that degrade the performance of traditional fixed-parameter automatic generation control (AGC). To address the resulting nonlinear dynamics, this paper proposes a frequency control strategy based on deep learning nonlinear model predictive control (DL-NMPC). Specifically, an attention-based bidirectional long short-term memory (Attention-BiLSTM) network is constructed to accurately model the variable-parameter frequency regulation dynamics of thermal units. Serving as the predictive core of the nonlinear model predictive control (NMPC) framework, this model forecasts grid frequency responses, while the whale optimization algorithm (WOA) is employed to solve the rolling optimization problem for optimal control actions. Simulation results demonstrate that the proposed strategy effectively resolves model mismatch issues under wide-load conditions and significantly outperforms traditional MPC in suppressing system frequency deviation. The scope of the current validation is confined to single-area thermal power systems, leaving the complex coupling dynamics, particularly the tie-line power exchange interactions and inter-area oscillations characteristic of multi-area interconnected grids, unaddressed in this study. The inherent dependence of the proposed data-driven model on the representativeness of the training dataset also implies that performance degradation may occur under extreme, unseen operating scenarios that significantly deviate from the historical distribution, posing potential challenges for robust control during rare fault events.