基于深度强化学习的中点钳位型三相有源电力滤波器直流电压与电容平衡协同控制OA
Coordinated control of DC voltage and capacitance balance in a neutral-point-clamped three-phase active power filter based on deep reinforcement learning
针对中点钳位型三相有源电力滤波器传统控制存在的直流母线电压调节与中点电压平衡耦合、比例积分(PI)控制器动态响应慢、参数整定依赖经验、抗干扰能力弱等问题,提出基于深度确定性策略梯度算法的直流侧电压协同控制策略.该策略构建适配有源电力滤波器高频开关特性与双控制目标的马尔可夫决策过程,设计复合型奖励函数,通过 L2 范数梯度化惩罚电压误差,实现直流母线电压稳定与分裂电容电压平衡的统一无差控制.Matlab/Simulink 仿真验证表明,在电容参数失配、直流电压启动、电网电压跌落等典型工况下,所提方法相较传统 PI 控制,直流母线电压调节时间缩短 57.1%,中点电容平衡调节时间缩短 55.6%,网侧电流总谐波畸变率降至2.57%,电网扰动下鲁棒性更强,为该类有源电力滤波器直流侧多目标协同控制提供了新路径.
To address the inherent flaws of the traditional control schemes for neutral-point-clamped three-phase active power filters,including the coupling between direct current bus voltage regulation and neutral-point voltage balancing,slow dynamic response of proportional-integral(PI)controllers,parameter tuning that relies on engineering experience,and poor disturbance rejection performance,this paper proposes a coordinated control strategy for direct current side voltage based on the deep deterministic policy gradient algorithm.This strategy constructs a Markov decision process tailored to the high-frequency switching characteristics and dual control objectives of the active power filter,devises a composite reward function,and imposes a gradient-based penalty on voltage errors via the second-order Euclidean norm,so as to achieve unified error-free control of direct current bus voltage stabilization and split capacitor voltage balancing.Simulation verification through Matlab/Simulink demonstrates that under typical operating conditions including capacitor parameter mismatch,direct current voltage startup,and grid voltage sags,compared with the traditional proportional-integral control,the proposed method reduces the direct current bus voltage regulation time by 57.1%,shortens the neutral-point capacitor balancing adjustment time by 55.6%,and lowers the total harmonic distortion of the grid-side current to 2.57%.The proposed method also exhibits stronger robustness under grid disturbances,which provides a novel approach for direct current side multi-objective coordinated control of this type of active power filters.
谢紫航;黄艳国
江西理工大学电气工程与自动化学院,江西 赣州 341000江西理工大学电气工程与自动化学院,江西 赣州 341000
有源电力滤波器中点钳位型直流侧电压协同控制深度强化学习
active power filterneutral-point-clampedcoordinated control of DC voltagedeep reinforcement learning
《电气技术》 2026 (9)
1-10,20,11
国家自然科学基金(72061016)江西省大学生创新创业训练计划项目(S202610407010)
评论