AI Summary & Key Takeaways
AI InsightThe results show that the THDI-based formula provides an interpretable initial estimate but cannot, on its own, account for field conditions such as phase asymmetry, load level, power factor, and voltage distortion.
📌 3 Key Takeaways
Abstract
In this study, the validity of the theoretical distortion factor based on total harmonic distortion of current (THDI) under industrial field conditions was evaluated, and low-complexity, data-driven correction models were developed to reduce formula error. The dataset was compiled from three-phase field measurements at 12 industrial measurement points. The measurement-based distortion factor was calculated from the PF/dPF ratio, while the theoretical distortion factor was calculated from the THDI-based equation. Since the phases at the same measurement point were not considered independent, the models were evaluated using a 12-cycle out-of-sample cross-validation method. The theoretical formula produced a mean absolute error of 0.015662, a root mean square error of 0.019681, and a mean bias of −0.009357. Among the compared models, restricted second-order regression produced the lowest error values, providing improvements of 31.83% and 15.70% in MAE and RMSE, respectively. Ridge regression, on the other hand, reduced the MAE to 0.012263 and brought the mean bias close to zero. However, the data-driven models did not reduce the maximum absolute error. At the measurement-point level, the Wilcoxon signed-rank test indicated lower paired errors after Ridge correction (P=0.0342), whereas the 95% bootstrap confidence interval for the mean paired MAE difference included zero. Taken together, these findings provide positive but limited evidence regarding the error-reduction potential of the Ridge correction. The results show that the THDI-based formula provides an interpretable initial estimate but cannot, on its own, account for field conditions such as phase asymmetry, load level, power factor, and voltage distortion. The proposed correction approach should be considered not as a universal formula but as an empirical calibration approach specific to the available field data.
Özet
In this study, the validity of the theoretical distortion factor based on total harmonic distortion of current (THDI) under industrial field conditions was evaluated, and low-complexity, data-driven correction models were developed to reduce formula error. The dataset was compiled from three-phase field measurements at 12 industrial measurement points. The measurement-based distortion factor was calculated from the PF/dPF ratio, while the theoretical distortion factor was calculated from the THDI-based equation. Since the phases at the same measurement point were not considered independent, the models were evaluated using a 12-cycle out-of-sample cross-validation method. The theoretical formula produced a mean absolute error of 0.015662, a root mean square error of 0.019681, and a mean bias of −0.009357. Among the compared models, restricted second-order regression produced the lowest error values, providing improvements of 31.83% and 15.70% in MAE and RMSE, respectively. Ridge regression, on the other hand, reduced the MAE to 0.012263 and brought the mean bias close to zero. However, the data-driven models did not reduce the maximum absolute error. At the measurement-point level, the Wilcoxon signed-rank test indicated lower paired errors after Ridge correction (P=0.0342), whereas the 95% bootstrap confidence interval for the mean paired MAE difference included zero. Taken together, these findings provide positive but limited evidence regarding the error-reduction potential of the Ridge correction. The results show that the THDI-based formula provides an interpretable initial estimate but cannot, on its own, account for field conditions such as phase asymmetry, load level, power factor, and voltage distortion. The proposed correction approach should be considered not as a universal formula but as an empirical calibration approach specific to the available field data.