Active Power Filters Applied to Smart Grids: Harmonic Content Estimation Based on Deep Neural Network

The increasing penetration of power electronics loads in smart grids inherently leads to severe concerns about power quality (PQ) disturbances. This chapter presents the principles of synthesizing control references for an active power filter (APF), which is placed in a smart grid comprising distortion loads, aiming at achieving PQ enhancement and compliance with standardized indexes. In addition, it is argued that the APF control system requires harmonic content identification to generate the targeted compensating currents. Thus, to achieve the disturbance recognition expected for synthesizing control references, harmonic analysis methods can be devised by automation tools and artificial intelligence (AI). For instance, it is demonstrated here that deep neural networks (DNN) can be used in the main pattern recognition stage, simplifying the harmonic content estimation process. The DNN-estimated harmonics are then directly used within the APF control system to compensate disturbances aiming at reducing the total harmonic distortion in a smart grid. Finally, this chapter also presents case studies using experimental load currents to depict the feasibility of the DNN-based method.