Multi-point harmonic contribution determination considering multicollinearity of measurement data

Abstract

Harmonic contribution determination is the basis for harmonic mitigation. To reach this goal, a well-known solution is to utilize the linear regression based methods. This paper scrutinizes this technique and points out a potential problem which may encounter in field applications: the multicollinearity between harmonic measurements. Multicollinearity could lead to unreliable estimated harmonic contributions of traditional methods. For example, linear loads may take on great harmonic contributions and the estimated contributions of harmonic sources are quite different from the exact values. Aiming at this problem, this paper proposes a harmonic contribution determination method based on elastic net regression. This method can reduce the singularity of the measurement matrix and filter out highly correlated data by introducing constraints to solve the ill-conditioned problem caused by multicollinearity. The methodology is verified through both simulation and field data. Comparative experiments and sensitivity studies are conducted to address some practical considerations. The results demonstrate the effectiveness and superiority of the proposed method in different scenarios.