Efficient learning approach using new combined fuzzy‑M5P model tree: experimental investigation of active power filters

This paper deals with the complete design and real-time implementation of a novel mixed control based on the pruned model tree (M5P) and collected datasets of a fuzzy logic controller. This combination aims to benefi t from both the decision tree rapidity and the fuzzy logic advantages. In harmonic mitigation systems with an active power fi lter, a strategy for identifying harmonic currents has a considerable infl uence on the quality and capacity of compensation. The proposed fuzzy-M5P model tree is assessed in the indirect current control identifi cation algorithm along with an eff ective comparison using the artifi cial neural network approach. The two learning methods are described and contrasted in an organized manner to evaluate their respective advantages in both steady state and dynamic state operating conditions. In compliance with IEEE std 519–1992 harmonic limits, an experimental setup was realized using dSPACE 1103 hardware to verify the excellent behavior of the system and to confi rm the eff ectiveness of the proposed M5P based control in terms of an almost unity power factor of 0.99, a low total harmonic distortion value of 3.07%, and satisfactory dynamic performances characterized by a fast response time of 100 ms.