Abstract: To address the strong inertia, significant time delay, and complex nonlinearity in boiler main steam temperature control systems, a multi-step predictive control method is proposed based on the integration of radial basis function neural network (RBFNN) and chaotic particle swarm optimization (CPSO). The RBFNN is employed to construct a nonlinear prediction model with online adaptive correction capability, enabling accurate forecasting of system output. Meanwhile, a chaotic mechanism is introduced to enhance the global search capability of the particle swarm optimization algorithm, which is used for rolling optimization to determine the optimal control input. Simulation results demonstrate that the proposed method achieves fast response, low overshoot, and strong disturbance rejection in main steam temperature regulation, effectively meeting practical control requirements.