Abstract: Compressed air energy storage (CAES) is an important technology for facilitating renewable energy integration. However, variations in load demand during charging can cause the compression subsystem to deviate from its high-efficiency operating region. To address this issue, this study proposes an economic model predictive control (EMPC) method based on an online incremental least squares support vector machine (OI-LSSVM). First, a dynamic physics-based simulation model of the CAES compression subsystem is developed, and an OI-LSSVM prediction model is constructed using incremental nonlinear autoregressive with exogenous inputs (NARX) features. An EMPC framework is then designed based on the prediction model, with constraints on control input magnitudes and increments. Optimal control inputs are determined to achieve coordinated optimization of multistage compressors. Simulation results demonstrate that the OI-LSSVM model provides adaptive updating capability and high prediction accuracy, with mean absolute percentage errors (MAPEs) of 0.55% and 0.67% for total power and useful power, respectively. Following a step increase in the total power setpoint from 116 MW to 117 MW, the proposed coordinated control method increases the overall efficiency of the compression subsystem from 88.39% to 88.55%, thereby improving its energy efficiency during dynamic operation.