Population-Based Genetic Algorithm for Loss and CO2 Emission Reduction in AC Microgrids with Distributed Energy Resources

Daniel Sanin-Villa; Luis Grisales-Noreña; Oscar Montoya

Abstract:

Introduction.
The accelerated integration of distributed energy resources into alternating current microgrids has increased the complexity of operational planning, particularly in reducing electrical energy losses and carbon dioxide emissions associated with slack-node power supply. These challenges are intensified by the intermittent nature of photovoltaic generation and the dynamic operation of battery energy storage systems. Population-based metaheuristic optimization methods have shown strong potential for addressing such nonlinear and constrained problems. Nevertheless, quantitative, statistically robust comparisons among algorithms remain limited, especially when technical and environmental objectives are treated independently.

Methods.
An intelligent energy management framework for AC microgrids is assessed, where the coordinated dispatch of photovoltaic units, battery energy storage systems, and D-STATCOMs is formulated as a nonlinear optimization problem. Two independent mono-objective functions are considered: minimizing total daily energy losses and minimizing total daily CO₂ emissions. A master–slave structure is adopted, in which candidate solutions are generated by population-based algorithms and evaluated using hourly power flow via successive approximations. The Population-Based Genetic Algorithm and Particle Swarm Optimization are implemented under identical tuning parameters and operational constraints, including nodal voltage limits, line loading limits, power balance equations, and battery state-of-charge bounds. The framework is tested on an IEEE 33-node AC microgrid with distributed energy resources over a 24-hour horizon with 30-minute resolution. To ensure statistical significance, each algorithm is executed 100 independent times for each objective function and operating mode.

Results.

The performance of the Population-Based Genetic Algorithm and Particle Swarm Optimization was assessed through 100 independent executions for each objective function and operating mode, enabling a statistically consistent comparison.

Under grid-connected operation, loss-minimization results show that the Genetic Algorithm achieves a best solution of 1,825.6 kW and a worst solution of 1,834.4 kW, with a mean of 1,829.6 kW and a standard deviation of 0.0965%. In contrast, PSO attains a best solution of 2,515.7 kW and a worst solution of 2,868.0 kW, yielding a mean of 2,673.9 kW and a significantly higher standard deviation of 3.2277%. The average computational time per execution is 150.0 s for the Genetic Algorithm and 216.1 s for PSO, resulting in total runtimes of 4.17 h and 6.00 h for the complete statistical assessment, respectively.

For CO₂ emission minimization in grid-connected mode, the Genetic Algorithm produces best and worst solutions of 13.5855 tCO₂ and 13.6112 tCO₂, with a mean value of 13.5989 tCO₂ and a standard deviation of 0.0442%. PSO yields a best solution of 15.1910 tCO₂ and a worst solution of 16.2613 tCO₂, with a mean of 15.7686 tCO₂ and a standard deviation of 1.5987%. The corresponding average execution times are 148.0 s for the Genetic Algorithm and 218.1 s for PSO, indicating a lower computational burden for the GA under these operating conditions.

In islanded operation, similar trends are observed. For loss minimization, the Genetic Algorithm reaches a best solution of 1,857.3 kW and a worst solution of 1,868.2 kW, with a mean of 1,862.4 kW and a standard deviation of 0.1347%. PSO presents a best value of 3,030.0 kW and a worst value of 3,320.0 kW, with a mean of 3,148.5 kW and a standard deviation of 1.5778%. Due to the increased computational complexity of islanded operation, the average execution time rises to 1,093.5 s for the Genetic Algorithm and 665.7 s for PSO, corresponding to total runtimes of 30.37 h and 18.49 h for the 100-run analysis.

For CO₂ emission minimization in islanded mode, the Genetic Algorithm achieves best and worst solutions of 22.5306 tCO₂ and 22.5939 tCO₂, with a mean of 22.5630 tCO₂ and a standard deviation of 0.0526%. PSO attains a best solution of 25.6710 tCO₂ and a worst solution of 27.9578 tCO₂, resulting in a mean of 26.8340 tCO₂ and a standard deviation of 1.6421%. The average execution times are 1,075.7 s for the Genetic Algorithm and 680.6 s for PSO, confirming that GA maintains lower variability and tighter convergence despite increased computational effort.

Conclusions.
The results demonstrate that both GA and PSO can reduce energy losses and CO₂ emissions in AC microgrids through coordinated management of distributed energy resources. However, based on extensive statistical testing with 100 independent runs per objective, the Population-Based Genetic Algorithm consistently achieves superior solution quality, significantly lower variability, and more stable convergence behavior than PSO. These findings provide statistically grounded evidence supporting the suitability of GA-based strategies for sustainable and reliable microgrid energy management.