OPTIMIZATION OF 20 KV FEEDER NETWORKS USING THE GAUSS-SEIDEL METHOD TO REDUCE LOSSES AT PLN UP3 SEMARANG

Abstract

The rising demand for electrical

energy, driven by population growth and

technological advancements, presents

challenges for PT PLN (Persero), Indonesia's

main electricity provider. One key issue is

reducing power losses in the distribution

network, which affects both system

efficiency and company revenue. This study

explores the use of the Gauss-Seidel method

for power flow analysis on the 20 kV

distribution network at PLN UP3 Semarang,

aiming to decrease losses and improve

revenue. Using data on line impedance and

load from PLN UP3 Semarang, the Gauss-

Seidel method is applied via a Python script

in Google Colab. The findings show that this

method effectively reduces network losses,

with potential financial benefits for PLN UP3

Semarang. This research also lays the

groundwork for future network optimization

strategies and contributes to the field of

power flow analysis. The study is focused on

the 20 kV network and does not compare the

Gauss-Seidel method with other approaches.

Growing demand for electricity and

the need for reliable distribution motivate

continuous improvement of power-flow

analysis and loss-reduction strategies in

Indonesia’s medium-voltage networks. This

paper applies the classical Gauss–Seidel (GS)

load-flow to the 20 kV feeders of PLN UP3

Semarang using field parameters (line R/X,

substation data, and aggregated loads) and an

open computational workflow in

Python/Google Colab. We build a Ybus

model, adopt the per-unit system, and

implement standard GS updates for PQ buses

with a practical convergence tolerance. The

study evaluates baseline conditions and

several optimization scenarios (e.g., modest

R/X adjustments reflecting conductor

upgrades, improved feeder balancing, and

initial-voltage tuning). Results show

consistent reduction of technical losses

across representative ULPs and at the UP3

level; monthly loss percentages also trend

downward during the observation horizon.

Voltage profiles improve at non-slack buses

while remaining within typical planning

limits. The analysis highlights how low-

complexity, data-driven GS studies can

support day-to-day planning decisions for

feeder reconfiguration and targeted

reinforcement. We discuss implementation

limits (data quality, simplifications, and

scenario dependence) and outline follow-up

steps, including PV-bus modeling,

comparison with Newton–Raphson, and

integration with economic screening curves.

The findings strengthen the case for using

GS-based what-if analyses as a lightweight

decision aid for utilities operating medium-

voltage distribution networks.