Deep-learning-based Fast System-level Harmonic Control Strategy for Multi-bus Voltages Detected APF in Distribution Systems

Conventional linearized-model-based system-level control strategy of active power filter (APF) has poor dynamic control performance when load changing happens. Therefore, a four-layer neural network is built to learn the convergence behavior of the linearized system-level harmonic mitigation model. Then, a deep-learning-based control strategy is proposed to achieve fast mitigation of multi-bus harmonic voltages by single APF. Finally, an eight-bus system with distributed harmonic loads is built in simulation. Simulation results proves the good dynamic performance of the proposed method. Moreover, compared with conventional implementation of deep learning method in system-level harmonic control, the proposed method benefits in lower data demand and simplified training process.