A new method based on a multiperiod optimization strategy is proposed to address the limited aero-engine acceleration performance across a broad flight envelope due to the traditional point-by-point optimization strategy for acceleration control law. First, the acceleration control law optimization method based on a multiperiod optimization strategy is introduced. This method takes multiperiod control variables as decision variables, aiming to solve the acceleration control law from a global optimization perspective and directly target the objective. Additionally, the mathematical model of the acceleration control law optimization problem is derived, demonstrating that this optimization problem can be transformed into a quadratic programing (QP) problem. Second, the design method for the engine broad flight envelope acceleration control plan is studied by isotherm correction to apply the acceleration control law to the broad flight envelope. Finally, simulation analysis indicates that under the ground standard condition (H = 0 km, Ma = 0, Tt2 = 288.15 K), the multiperiod optimization strategy reduces the acceleration adjustment time by 9.431% compared to the traditional acceleration control law optimization method. The multiperiod optimization strategy significantly improves the engine acceleration performance and the compressor stability margin. The study also validates the applicability and feasibility of the broad flight envelope acceleration control method based on isotherm correction.

To address the steady-state control error issue in traditional deep reinforcement learning for command-tracking control tasks, a novel weighted MAX Bellman operator is proposed in this paper. The weighted MAX Bellman operator is defined, and its monotonicity and contractivity are proven to ensure iterative convergence. By replacing the traditional action-value function with the weighted MAX value function of its fixed points, the operator guides the agent to prioritize the optimization of steady-state performance. Finally, a deep reinforcement learning controller is designed using the double-delay deep deterministic policy gradient algorithm (TD3) as the baseline algorithm, and digital simulation verification is conducted on a component-level model of a small-bypass-ratio twin-spool mixed-flow turbofan engine. The results show that the proposed weighted MAX double-delay deep deterministic policy gradient algorithm (W-MAX TD3) achieves convergence speeds comparable to the baseline algorithm, reduces the median steady-state control error to 0.46%, and maintains the error within 2% even with a small-scale neural network. This represents an approximately 90% reduction in error compared to the traditional TD3 algorithm, while the dynamic response time remains stable at 3.5 seconds without significant degradation. The method proposed in this paper significantly improves steady-state accuracy without compromising dynamic performance. It is suitable for computationally constrained onboard control scenarios and provides theoretical support and technical references for the engineering application of deep reinforcement learning in the intelligent control of aircraft engines. Keywords: Deep Reinforcement Learning; Gas Turbines; Aero-engine Control; Steady-state Control Precision; Intelligent Control.