Fuzzy logic based hybrid load balancing for solar–grid supply with tiered load shedding: a simulation study

Abstract

Hybrid installations that combine a solar inverter and battery with the utility grid are increasingly common in Uganda, yet they are typically managed by threshold-based automatic transfer switches (ATS) that ignore solar availability, grid quality, and load criticality. This paper presents the design, modelling, and simulation based evaluation of a multi-input fuzzy logic based hybrid load balancing controller. The framework fuses four inputs (solar output, battery state of charge (SOC), grid status, and load demand) through a Mamdani-type fuzzy inference system to produce two coordinated outputs: a source-priority signal and a tiered load-shedding signal spanning three load classes (critical, flexible, and non-essential). The study is entirely simulation-based. The system was implemented in Python at one-minute resolution over a synthetic but representative 24-hour scenario featuring two grid outages, a brownout, and cloud-induced solar variability, and was benchmarked against a conventional threshold-based transfer switch, an ATS with critical-load priority, and a crisp rule-based tiered controller; it was further tested on seven stress cases and a 100-run Monte Carlo evaluation. In the head line scenario the fuzzy controller sustained 100% availability of critical load against 91.9% for the baseline, which suffered 67 minutes of complete blackout; raised solar utilisation from 47.1% to 100%; cut grid energy from 13.43 kWh to 7.95 kWh (40.8%); and reduced source-switching events from four to two. These gains came from deliberately curtailing all non-essential load and 4.6% of flexible load, so they are not a pure effi ciency gain: total unserved energy rose from 1.84 to 3.18 kWh, although a criticality-weighted reliability index still favoured the fuzzy controller (96.4% versus 92.6%). Broader testing confirmed the solar-utilisation and grid-energy advantages over the threshold baseline, but showed that an ATS with simple critical-load priority protected critical load at least as well under stress, identifying rule-base tuning as the main need. The results support a simulation-based proof of concept. A real-time embedded demonstration, in which the controller and plant model ran on a Raspberry Pi 5 with an external touch display using emulated inputs, reproduced the simulated behaviour; testing with live sensors and relays and field validation remain future work.