Net saving improvement of capacitor banks in power distribution systems by increasing daily size switching number: A comparative result analysis by artificial intelligence

Abstract

Abstract This paper studies the effect of the number of switching (NOS) per day of capacitor banks on loss reduction in radial distribution systems. To this aim, the daytime (more precisely, 24 h) is divided into different numbers of time segments (equal to the same NOS) for capacitors’ size switching. The resulting non‐linear programming with discontinuous derivatives (called DNLP) model is solved subject to related constraints. The results reveal the impact of hourly switching of capacitor banks on further loss reduction (namely 118.4435, 83.7856, and 101.738 MWh for three IEEE systems) and higher net savings (i.e. k$5.6067, k$4.2772, and k$5.3542 for the same systems) of radial distribution systems compared to daily switching. Then, the hyper‐tuned Random Forest model is trained based on the IEEE 69‐bus network, fine‐tuned by the IEEE 10‐bus network, and fitted by the IEEE 33‐bus network to have an intelligent multi‐classification task with the highest accuracy. Numerical simulation, in both classic and intelligent parts, is presented to demonstrate the performance of DeepOptaCap. For the final step, DeepOptaCast is compared to other intelligent models of Light Gradient Boosting Method (LGBM), Decision Tree, and XGBoost, regarding KPIs of mean absolute percentage error, root mean squared percentage error, mean absolute error, root mean squared error, and coefficient of determination to demonstrate the model's superiority.

Keywords

• distribution networks planning
• energy management
• engineering optimization
• optimal power flow
• optimal scheduling
• power factor correction