Researchers have identified significant differences in the energy consumption of four widely used nature-inspired optimization algorithms, with Differential Evolution (DE) emerging as the most energy-efficient option in a controlled experimental setting. The study, published in Green Technology, Resilience, and Sustainability, provides a critical analysis of the energy usage and corresponding carbon footprint of Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Differential Evolution (DE), and Artificial Bee Colony (ABC) algorithms. By employing Microsoft Joulemeter to measure runtime energy consumption, the team demonstrated that each algorithm exhibits distinct energy behaviors when solving the same optimization task. These findings are particularly relevant for software engineers and practitioners who prioritize energy efficiency in the selection of optimization tools for real-world applications.
The research addresses a gap in the existing literature, which has largely focused on the runtime of these algorithms rather than their environmental impact. Nature-Inspired Optimization (NIO) algorithms are population-based metaheuristics that mimic diverse phenomena in nature, such as biological evolution and swarm intelligence. They are prevalent in various fields, including telecommunications, image processing, engineering design, and vehicle routing, due to their simplicity, flexibility, and effectiveness in circumventing local optima. However, the proliferation of mobile and IoT devices has made software energy consumption a crucial factor in the design of high-performance and mobile computing systems.
The study utilized a Life Cycle Assessment (LCA) methodology for ICT Goods, Networks, and Services as its macro framework. This systematic analytical method assesses the potential environmental effects of ICT services in a consistent and transparent manner, covering stages from raw material acquisition to end-of-life treatment. In this specific context, the focus was on the use stage, where the algorithms’ impact on energy use is most directly observable. The micro methodology involved implementing the four NIO algorithms in MATLAB and measuring their energy consumption using Microsoft Joulemeter, a tool capable of tracking energy used by running applications and individual hardware resources such as the CPU, monitor, and disk.
To ensure a fair comparison, the researchers designed a rigorous experimental setup using a commonly used benchmark function known as the Sphere Function. This function is defined by the sum of the squares of the variables, with a global minimum of zero at the origin. The search space for each variable was set between -5.12 and 5.12, with a dimension size of 50. The population size was set to 10 times the dimension size, and the maximum number of iterations was 100 times the dimension size. Each algorithm was executed ten times separately to capture consistent measurements and avoid outliers, with the optimal value to reach set at 10 to the power of negative 8.
The results indicated that the DE algorithm consumed the least energy, followed by PSO and GA, while the ABC algorithm had the highest energy consumption. To facilitate a direct comparison, the energy consumption was normalized for one second of runtime. The study found that running the ABC algorithm consumes almost four times more energy compared to the DE algorithm. In contrast, the energy consumed by the GA algorithm is approximately double that of the DE algorithm, while PSO consumes comparatively lower energy than both ABC and GA. These differences highlight the significant impact that algorithm choice can have on the overall energy footprint of computational tasks.
In addition to energy consumption, the researchers calculated the carbon footprint of each algorithm based on the UK DEFRA guide for greenhouse gas conversion factors. The recorded energy consumption in kilowatt-hours was converted into equivalent CO2 emissions using an emission factor of 0.25319 kgCO2e per kWh. The analysis confirmed that the DE algorithm has the lowest carbon footprint, followed by PSO and GA, with the ABC algorithm exhibiting the highest carbon footprint among the four. This correlation between energy usage and carbon emissions underscores the environmental implications of selecting less efficient optimization algorithms for large-scale computational projects.
To verify the statistical significance of these differences, the team conducted a one-way Analysis of Variance (ANOVA) test. The null hypothesis stated that the average energy consumed by each algorithm was equal, while the alternative hypothesis proposed that the average energy consumption was significantly different. Using Excel Data Analysis ToolPak, the test was performed at a significance level of 0.05. The resulting p-value was 5.85E-37, which is substantially less than the significance level. Consequently, the null hypothesis was rejected, leading to the conclusion that the average energy consumed by each algorithm is significantly different from the others.
The study also examined the relationship between hardware and application energy consumption. It was observed that the energy consumption of the hardware running the application is much higher compared to the energy consumed by the application itself for each algorithm. This finding suggests that while optimizing the algorithm is important, the interaction and processes between hardware and software also play a critical role in overall energy efficiency. The researchers noted that different hardware specifications would bring about different results, emphasizing the need for further investigation into how software can be optimized to reduce the load on hardware resources.
Limitations of the study include the fact that only four types of NIO algorithms were investigated, and the experiments were conducted on a specific laptop with defined specifications. The authors acknowledge that there are more than a hundred existing NIO algorithms and their variants in the literature, and thus, the results cannot be generalized to all possible optimization algorithms. Furthermore, the specific hardware used in the study means that the results may vary on different devices. The researchers suggest that the experimental procedures could be replicated for other NIO algorithms and on different CPU architectures, such as AMD and ARM, to identify broader trends in energy efficiency.
Future work proposed by the researchers includes analyzing more NIO algorithms and their variants to identify the top ten greenest algorithms. They also suggest considering more benchmark functions and real-life optimization problems to evaluate energy consumption in more complex scenarios. Additionally, investigating the tuning and controlling of algorithm-dependent parameters could help maximize performance while reducing energy consumption. The study aims to raise awareness of the energy efficiency issues associated with NIO algorithms and encourage further research in this field, ultimately helping developers choose the most appropriate and green algorithm for specific domain problems when energy efficiency is a top priority.
The study was conducted by Mohammad Newaj Jamil and Ah-Lian Kor, with the latter identified by the ORCID 0000-0001-9514-3773. The research was published in the journal Green Technology, Resilience, and Sustainability, specifically in Volume 2, article number 1, on January 27, 2022. The article is designated as original research and is available under an open access license, allowing full access to the content without subscription barriers. The publication has recorded 4,213 accesses. The authors positioned this work as a response to the growing concern over the environmental impact of Information and Communication Technology (ICT), noting that the sector is accountable for two percent of global carbon emissions. While other sectors such as health, transportation, and education are responsible for the remaining 98 percent, the authors argue that the ICT sector can play a crucial role in reducing global carbon emissions by minimizing the carbon footprints of its products and services. This context highlights the importance of assessing the energy efficiency of software components, including optimization algorithms, as part of broader sustainability efforts.
The selection of the four specific algorithms—Genetic Algorithm, Particle Swarm Optimization, Differential Evolution, and Artificial Bee Colony—was driven by their wide-ranging application areas. The authors noted that these algorithms are prevalent in fields such as telecommunications, image processing, engineering design, and vehicle routing. Each algorithm has distinct characteristics and limitations that influence their performance. For instance, the Genetic Algorithm is based on Darwinian evolution and uses crossover, mutation, and selection operators, but it suffers from premature convergence that can lead to the loss of alleles. Particle Swarm Optimization is derived from the swarm intelligence of flocking birds and schooling fish, where each particle contains its own velocity and position. However, PSO easily falls into local optima in high-dimensional spaces and has a low convergence rate. Differential Evolution is formulated on vector population evolution, using differential crossover and mutation operators, but it exhibits unstable convergence and can easily fall into regional optima. The Artificial Bee Colony algorithm is elicited from the food searching behavior of honey bee swarms, comprising food sources, employee foragers, and unemployed foragers. Its primary limitation is slow convergence and susceptibility to premature convergence.
The experimental design included specific parameters to ensure consistency across the four algorithms. The dimension size was set to 50, with a population size of 500 (10 times the dimension size) and a maximum iteration count of 5,000 (100 times the dimension size). The optimal value to reach was set at 10 to the power of negative 8. The search space for each variable was defined between -5.12 and 5.12. These parameters were kept identical for all algorithms to analyze the measurements’ consistency and avoid outliers. The use of the Sphere Function as a benchmark is significant because it is a commonly used test function in optimization research, providing a standardized metric for comparison. The authors emphasized that different hardware specifications would bring about different results, which is why all experiments were conducted on a single laptop with defined specifications. This controlled environment helps isolate the energy consumption attributable to the algorithmic logic rather than hardware variability. The manual calibration of the Microsoft Joulemeter was necessary because the Windows 8.1 Pro operating system lacked support for automatic calibration, a detail that underscores the technical rigor required to obtain accurate energy measurements in this specific setup.
The study also addressed the ethical and validity challenges associated with the research. The authors categorized potential validity issues into four areas: conclusion validity, internal validity, construct validity, and external validity. Regarding internal validity, the authors noted that while other factors such as specific laptop versions or uncontrollable system processes could contribute to variations, the consistent execution of each algorithm ten times helped minimize these effects. The measured results were described as quite consistent and reliable, supported by the proven accuracy of the energy measurement tool. Construct validity was addressed by implementing all four algorithms in the same programming language, MATLAB, and using the same benchmark function, which allowed for a direct comparison of energy consumption for the same task execution. External validity was acknowledged as a limitation, as the results cannot be generalized to all NIO algorithms or different hardware devices. The authors suggested that the experimental procedures could be replicated for other algorithms and on different CPU architectures, such as AMD and ARM, to identify broader trends. This replication potential is a key strength of the study, as it provides a clear methodology that other researchers can follow to extend the analysis to a wider range of optimization algorithms and hardware configurations.
Subject of Research: Energy consumption and carbon footprint analysis of nature-inspired optimization algorithms
Article Title: Analyzing energy consumption of nature-inspired optimization algorithms
Article References: Jamil, M. N., & Kor, A.-L. (2022). Analyzing energy consumption of nature-inspired optimization algorithms. Green Technology, Resilience, and Sustainability, 2(1), Article 1. https://doi.org/10.1007/s44173-021-00001-9
Image Credits: AI Generated
DOI: 10.1007/s44173-021-00001-9
Keywords: nature-inspired optimization, energy efficiency, carbon footprint, Differential Evolution, Genetic Algorithm, Particle Swarm Optimization, Artificial Bee Colony, green computing, MATLAB, Microsoft Joulemeter, Analyzing, energy
News Source: Denise Maddox. (November 9, 2026). Analyzing energy consumption of nature-inspired optimization algorithms. Scienmag.
