Atom Search Optimization: a comprehensive review of its variants, applications, and future directions

Inspiration and principle

The theoretical foundation of ASO is inspired by molecular dynamics, a field that studies the structure of matter and atomic interactions at the microscopic scale. ASO employs a search mechanism based on the interaction and constraint forces between pairs of atoms (Zhao, Wang & Zhang, 2019a). Interaction forces, encompassing both repulsive and attractive components, mimic the forces that bind atoms together. Constraint forces, on the other hand, are used to maintain fixed distances between atoms connected by covalent bonds. Throughout the iterative process, atoms interact via these forces. When the distance between two atoms is less than the equilibrium distance, the force is attractive; conversely, it becomes repulsive when it exceeds the equilibrium distance. Additionally, a constraint force acts between each atom and the atom with the highest fitness (Zhao, Wang & Zhang, 2019a).

Mathematical model

Three key mathematical components are defined in the atomic system: interaction force, geometric constraint force, and atomic motion. The details of these components are provided below.

Framework of ASO algorithm

The ASO optimization process begins by generating an initial random set of solutions. During each iteration, atoms adjust their positions and velocities, with the position of the best-performing atom also being updated. The acceleration of each atom is determined by two components: the interaction force from the Lennard-Jones potential (which combines attractive and repulsive forces) and the constraint force from the bond-length potential (which represents a weighted difference between the atom’s position and the position of the best atom). These updates and calculations are repeated iteratively until the termination criteria are satisfied. Lastly, the position and fitness value of the best atom approximates the global optimum. The pseudo-code for the ASO algorithm is outlined in Algorithm 1.

Complexity analysis of ASO algorithm

The computational complexity of the ASO algorithm depends on the problem dimensionality (D), atom population size (N), and maximum iteration count (T). The primary computational costs arise from the initialization phase, fitness evaluations, memory management, position updates, and acceleration updates. Consequently, the overall computational complexity of the ASO algorithm can be expressed as follows:

(17) $$\begin{array}{c}O(\mathrm{A}\mathrm{S}\mathrm{O})=O(\mathrm{P}\mathrm{r}\mathrm{o}\mathrm{b}\mathrm{l}\mathrm{e}\mathrm{m}\mathrm{d}\mathrm{e}\mathrm{f}\mathrm{i}\mathrm{n}\mathrm{i}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n})+O(\mathrm{I}\mathrm{n}\mathrm{i}\mathrm{t}\mathrm{i}\mathrm{a}\mathrm{l}\mathrm{i}\mathrm{z}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n})+O(\mathrm{F}\mathrm{u}\mathrm{n}\mathrm{c}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}\mathrm{e}\mathrm{v}\mathrm{a}\mathrm{l}\mathrm{u}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n})\hfill \\ +O(\mathrm{M}\mathrm{e}\mathrm{m}\mathrm{o}\mathrm{r}\mathrm{y}\mathrm{s}\mathrm{a}\mathrm{v}\mathrm{i}\mathrm{n}\mathrm{g})+O(\mathrm{P}\mathrm{o}\mathrm{s}\mathrm{i}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}\mathrm{u}\mathrm{p}\mathrm{d}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{n}\mathrm{g})+O(\mathrm{A}\mathrm{c}\mathrm{c}\mathrm{e}\mathrm{l}\mathrm{e}\mathrm{r}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}\mathrm{u}\mathrm{p}\mathrm{d}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{n}\mathrm{g})\hfill \\ O(\mathrm{A}\mathrm{S}\mathrm{O})=O(1)+O(ND)+O(TCN)+O(TN)+O(TND)+O(TND)\hfill \\ O(\mathrm{A}\mathrm{S}\mathrm{O})=O(1+ND+TCN+TN+TND+TND)\hfill \\ O(\mathrm{A}\mathrm{S}\mathrm{O})\cong O(TND+TCN).\hfill \end{array}$$

Parameter sensitivity

The ASO algorithm is notably simple to implement and requires minimal parameter tuning. Aside from the standard parameters such as maximum iterations, atom population size, and problem dimension, only the upper and lower bounds need to be defined. For the function $F^{\mathrm{\prime }}$, setting $g_0=1.1$, and $u=1.24$ generally produces good results. The depth and multiplier weights, represented by $\alpha$ and $\beta$, respectively, can be empirically adjusted within 0 to 100 for $\alpha$ and 0 to 1 for $\beta$ (Zhao, Wang & Zhang, 2019a). However, extensive testing suggests optimal performance is typically achieved within the narrower ranges of $40\le \alpha \le 60$ and $0.1\le \beta \le 0.3$ (Zhao, Wang & Zhang, 2019a). These ranges serve as a general guideline, as specific problems may require fine-tuning these parameters. Nevertheless, setting $\alpha =50$ and $\beta =0.2$ offers a reasonable starting point for many applications. Researchers can further adjust these parameter values based on the unique characteristics of their optimization problem to enhance performance.

ASO against state-of-the-art algorithms

This subsection provides a comprehensive assessment of the ASO algorithm, comparing it with other state-of-the-art algorithms, as summarized in Table 2. As shown in the table, ASO exhibits distinct characteristics compared to other algorithms regarding the number of parameters and complexity. However, it also shares specific strengths and weaknesses with these algorithms. The analysis highlights that the basic ASO requires enhancements to improve its capabilities, address its limitations, and achieve superior performance compared to other algorithms.

Limitation of ASO algorithm

ASO demonstrates robust optimization capabilities and has been applied to various engineering challenges. However, it has notable limitations. The velocity update equation in ASO relies exclusively on individual atom velocities and accelerations, which can reduce population diversity and limit information exchange by neglecting the influence of atom positions (Sun et al., 2021a; Wang et al., 2022b). Furthermore, the attractive force in ASO can cause atoms to cluster prematurely, leading to early convergence. Conversely, the repulsive force may be insufficient to help atoms escape local optima in later stages (Joshi et al., 2024). This premature convergence limits exploration, while the inability to escape local optima affects exploitation, ultimately limiting ASO’s effectiveness in solving complex optimization problems. Various variants have been proposed to address these challenges and enhance ASO’s performance, as detailed in the following section.

Modified versions of atom search optimization

Like many algorithms, ASO can face challenges in achieving global optima, such as slow convergence, local optima traps, and an imbalance between exploration and exploitation. To overcome these limitations, researchers have introduced several modifications to the ASO search process. Additionally, ASO has been adapted to handle different search spaces, extending its applicability beyond its original design for continuous domains. This subsection reviews the modified ASO variants, categorizing them based on the types of modifications, with each category discussed in its respective subsection.

Hybridized versions of atom search optimization

A hybrid algorithm combines multiple algorithms to seek optimal solutions for a given problem (Bouaouda & Sayouti, 2022). Researchers across various disciplines worldwide have increasingly recognized ASO. Consequently, diverse applications and hybrid extensions of the basic ASO algorithm have been developed.

• Hybridization with Particle Swarm Optimization

Zhao et al. (2021) developed the AASOPSO algorithm, which combines Atom Search ASO with PSO to address ASO’s issues with local optima and search efficiency. AASOPSO enhances the exploitation phase with PSO’s cognitive components and uses adaptive acceleration coefficients to balance exploration and exploitation. Additionally, it adjusts each atom’s force constant based on fitness feedback. The algorithm outperformed other optimizers in benchmark tests and in optimizing the no-load PID design of a hydro-turbine governor, demonstrating superior global search performance.

Wang et al. (2022b) proposed the Improved ASO (IASO) to overcome the limitations of the original ASO, such as local optima and lower accuracy. By integrating the velocity update mechanism from PSO, IASO achieved faster convergence and higher precision across 23 benchmark functions. Applied to enhance the maximum likelihood (ML) estimator for the direction of arrival (DOA), IASO demonstrated superior performance in convergence speed, iteration count, and lower root mean square error (RMSE) compared to ASO, SCA, GA, and PSO, showing promise for improving accuracy and reducing complexity in multidimensional optimization tasks.

Izci, Ekinci & Hussien (2023) developed the hybrid Atom Search Particle Swarm Optimization (h-ASPSO) algorithm, merging ASO and PSO to improve optimization by balancing exploration and exploitation. h-ASPSO demonstrated enhanced search efficiency and time-domain performance in complex tasks, such as designing PID controllers and optimizing wind turbine systems. It outperformed the original ASO in convergence speed and solution quality and competed well with other advanced methods while maintaining computational efficiency.

Al Sumarmad et al. (2023) proposed a coordinated power management strategy for battery energy storage (BES) in renewable energy (RE) integrated DC microgrids (MGs) to address power and voltage imbalances from intermittent RE sources. Their system employs a hybrid optimization technique, combining ASO and PSO, to optimize FOPI controllers. This approach enhances the DC network’s control response and voltage regulation, improving BES performance for stable and efficient microgrid operation.

• Hybridization with simulated annealing

Eker et al. (2021) developed the hybrid Atom Search Optimization (hASO-SA) algorithm by integrating simulated annealing (SA) with ASO. This approach enhances ASO’s search capabilities for nonlinear and linear optimization tasks. Tested on various benchmark functions and applied to practical problems like multilayer perceptron training and PID controller design, hASO-SA outperformed other metaheuristic methods and SA-based hybrids in accuracy and robustness. It showed notable improvements in transient and frequency responses, highlighting its effectiveness across various optimization problems.

Izci (2022) proposed an improved hybrid ASO algorithm by incorporating SA. This variant was designed to optimize power system stabilizers in a single-machine infinite-bus power system. It demonstrated superior performance on benchmark functions compared to Genetic Algorithm, Particle Swarm Optimization, and the original ASO. Additionally, it effectively optimized a power system stabilizer damping controller in real-world applications, outperforming methods such as the Sine-Cosine Algorithm (SCA) and Symbiotic Organisms Search.

• Hybridization with Sine Cosine Algorithm

Abd Elaziz et al. (2019) developed a hybrid algorithm named ASOSCA, combining ASO with the SCA for automatic clustering. This method optimizes centroids for complex tasks like image clustering and bioinformatics. By integrating SCA, ASOSCA enhances clustering performance and outperforms other metaheuristics on multiple datasets using various cluster validity indexes.

Singh & Kaushik (2021b) developed ASSCA for multi-focus image fusion. This hybrid approach combines ASO with the Sine Cosine Algorithm to optimize the fusion factor in a discrete wavelet transform (DWT) framework. By using Renyi entropy to weigh source images, ASSCA enhances fusion accuracy. The method demonstrated superior performance in metrics like mutual information, peak signal-to-noise ratio, and correlation coefficient, outperforming traditional image fusion techniques.

In other work, Singh & Kaushik (2021a) introduced WeAbDeepCNN, an advanced image fusion technique that tackles contrast reduction, block artifacts, and artificial edges. This method combines a weighted average model with a deep convolutional neural network (DCNN) optimized using the Atom Search Sine Cosine Algorithm (ASSCA). The technique achieves high mutual information, peak signal-to-noise ratio, and low root mean square error, demonstrating superior image clarity and quality compared to existing methods.

Furthermore, Singh & Kaushik (2022) developed a multimodal medical image fusion technique using the Atom Search Sine Cosine Algorithm (ASSCA) with a DCNN. The method integrates Dual-Tree Complex Wavelet Transform (DTCWT) for sub-band decomposition and employs a Renyi entropy-based weighted fusion model. Guided by ASSCA-DCNN, the technique achieved significant improvements in image quality, as demonstrated by higher peak signal-to-noise ratio (PSNR), mutual information (MI), and lower root mean square error (RMSE), highlighting its potential in advancing multimodal medical image fusion.

Zhou et al. (2023) developed the ASCASO algorithm, a hybrid version of the ASO algorithm for better PV model parameter estimation. ASCASO integrates an anti-sine-cosine mechanism, combining sine-cosine algorithm principles with a mutation strategy from linear population size reduction adaptive differential evolution (LSHADE). This approach improves population diversity and communication through arcsine and arccosine functions. Applied to various PV models, ASCASO outperformed nine other algorithms in accuracy and reliability, demonstrating its effectiveness in optimizing PV model parameters.

• Hybridization with Squirrel Search Algorithm

Borhade & Nagmode (2020) introduced a Modified ASO (MASO) algorithm with a DRNN for epileptic seizure prediction. MASO uses EEG signals to optimize feature selection and DRNN training, enhancing accuracy and reducing computation time. The MASO, which integrates aspects of the Squirrel Search Algorithm and ASO, was tested on the CHB-MIT Scalp EEG dataset, achieving high performance with notable accuracy, sensitivity, and specificity. This approach offers fast convergence, low complexity, and robustness, making it effective for early seizure prediction.

Borhade et al. (2023) developed an Adaptive Exponential Squirrel Atom Search Optimization (Adaptive Exp-SASO) based DRNN for epileptic seizure prediction. This method improves prediction accuracy by using Gaussian filters to clean EEG signals, applying Fuzzy Information Gain (FIG) for feature selection, and optimizing DRNN weights with Adaptive Exp-SASO. The approach achieved 97.87% accuracy, 97.85% sensitivity, and 98.88% specificity, demonstrating strong performance in seizure prediction.

• Hybridization with Whale Optimization Algorithm

Halhalli, Sugave & Jagdale (2021) introduced a secure routing protocol for mobile ad hoc networks (MANETs) using the Atom Whale Optimization Algorithm (AWOA). This hybrid approach combines ASO and WOA to optimize route selection based on trust factors like encounter rate and cooperation frequency. By integrating these trust metrics into its fitness function, AWOA enhances routing security and performance. The protocol showed significant improvements over existing methods, with reduced end-to-end delay, higher packet delivery ratio, and increased throughput, making it more reliable and efficient for dynamic MANETs.

Singh & Kumar (2024) introduced CWASO-based Deep Stacked Autoencoder (Deep SAE) for crowd behavior recognition. This model uses a combination of ASO and Chaotic Whale Optimization (CWOA) to optimize feature extraction and selection, including statistical and visual features. Evaluated on two datasets, it achieved high precision (96.826%), sensitivity (96.790%), and specificity (99.395%) on dataset 1, demonstrating its effectiveness in crowd behavior recognition.

• Hybridization with Equilibrium Optimizer

Chattopadhyay et al. (2023) introduced the CEOAS algorithm, a hybrid feature selection method combining a Clustering-based Equilibrium Optimizer (EO) with ASO. This approach improves feature reduction and classification accuracy in high-dimensional machine-learning tasks. Applied to speech emotion recognition, CEOAS effectively extracted features like LPC and LPCC, achieving up to 98.72% recognition accuracy on datasets such as SAVEE, EmoDB, RAVDESS, and IEMOCAP, demonstrating ASO’s effectiveness in optimizing data processing for better machine learning results.

Maazalahi & Hosseini (2024) proposed a two-stage hybrid method for attack detection that combines ASO and Equilibrium Optimization (EO) for feature selection with K-means clustering and the Firefly Algorithm (FA) for classification. This approach, termed ASO-EO-FA-K-means, was tested on NSL-KDD, UNSW_NB15, and KDD_CUP99 datasets, achieving high accuracy (0.998, 0.995, and 0.995) and low error rates, outperforming other methods like PSO and Genetic Algorithms, with notable results on the NSL-KDD dataset.

• Hybridization with Multi-verse Optimizer

Maheswari, Siva & Priya (2023) introduced the Optimal Cluster-based Intrusion Detection System (OC-IDS), which uses a hybrid Multi-verse and Chaotic ASO (MCA) for data pre-processing and Chaotic Manta-ray Foraging Optimization (CMFO) for clustering. The system achieved 95.01% accuracy on datasets like KDD Cup’99, demonstrating the effectiveness of the hybrid method in cloud security.

Pitchandi, Nivaashini & Grace (2024) proposed an advanced IDS model that combines deep learning with optimization techniques, including Multi-verse-based Chaotic ASO (MCA) for preprocessing, PCA, and Chaotic Manta-ray Foraging Optimization (CMRFO) for feature selection. The ensemble model integrates SVM with CNN, LSTM, and GRU, achieving superior accuracy, precision, recall, and F-measure compared to traditional methods, as demonstrated with the Aegean Wi-Fi Intrusion Dataset (AWID).

• Hybridization with Grey Wolf Optimizer

Bhagat, Ramesh & Patil (2019) proposed a hybrid method for automatic epileptic seizure diagnosis using the Atom Grey Wolf Optimization (AGWO) algorithm, which combines ASO with GWO. This hybrid approach was designed to enhance the training efficiency of a deep-stacked autoencoder for classifying EEG signals. By leveraging the strengths of both ASO and GWO, the AGWO achieved impressive performance, including 94.103% accuracy, 91.98% sensitivity, and 97.344% specificity, demonstrating its potential for clinical seizure detection.

• Hybridization with Water Wave Optimization

Dabbu et al. (2022) developed a big data classification approach using a Spark architecture and the Water Atom Search Optimization (WASO) algorithm, which combines Water Wave Optimization (WWO) and ASO algorithm. The process starts with data partitioning via feature selection using Renyi entropy and Black Hole Entropy Fuzzy Clustering (BHEFC). These features are processed by a Deep Recurrent Neural Network (DeepRNN) optimized with WASO. The approach showed significant improvements in classification performance, with higher specificity, accuracy, sensitivity, and precision compared to methods like NN and Support SVM, proving effective for practical big data applications.

• Hybridization with Rider Optimization Algorithm

Shyjith, Maheswaran & Reshma (2021) proposed the Rider-ASO to improve secure and efficient multipath routing in WSNs. Combining ASO with the Rider Optimization Algorithm (ROA), Rider-ASO optimizes cluster head selection and secure routing. The algorithm achieved notable results: minimal delay of 0.009 s, maximum average residual energy of 0.5494 J, a packet delivery ratio of 97.82%, and a throughput rate of 96.07%, outperforming existing methods.

• Hybridization with Sunflower Optimization

Jadhav & Joshi (2021) introduced the Atom Search Sunflower Optimization (ASSFO) algorithm to enhance trust-based routing in IoT networks, addressing vulnerabilities like sinkhole attacks. By combining Sunflower Optimization (SFO) with the ASO algorithm, ASSFO leverages multiple trust factors to improve network security and performance. The method surpassed earlier approaches, achieving superior throughput, energy efficiency, trust, and packet delivery ratio (PDR) results.

• Hybridization with Bird Swarm Algorithm

Alhassan & Zainon (2022) introduced the Atom Taylor BSA-based DBN to enhance medical data classification. This hybrid method combines ASO with the Taylor Bird Swarm Algorithm (Taylor BSA) to optimize feature selection for a Deep Belief Network (DBN) classifier. The approach addresses missing values and redundant attributes, improving classification accuracy. Experimental results, especially using the Hungarian dataset, showed significant gains in specificity, sensitivity, and overall accuracy.

• Hybridization with Emperor Penguin Optimization

Angel & Jaya (2022) introduced a Hybrid Emperor Penguin Optimization (EPO) algorithm combined with the ASO algorithm to improve Wireless Sensor Networks (WSNs). This hybrid approach enhances EPO’s updating function, focusing on load balancing, security, and energy efficiency. Validated with the NS2 simulator, the method outperformed conventional EPO and ASO in metrics such as delivery ratio, network lifetime, energy consumption, throughput, and delay, providing a robust solution for optimizing WSN performance.

• Hybridization with Tree-Seed Algorithm

Almajed et al. (2023) developed a forecasting method for non-fungible token (NFT) prices using adaptive improved convolutional neural networks (AICNN) combined with the Tree-seed Chaotic ASO (TSC-ASO) algorithm. This approach incorporates Web3 Blockchain technology to enhance prediction accuracy and integrity. TSC-ASO refines model parameters to improve forecasting reliability. The method outperformed existing techniques, providing more accurate and insightful NFT price predictions in the digital asset market.

• Hybridization with Aquila Optimizer

Subramaniam & Singh (2023) developed a novel approach for allocating plug-in hybrid electric vehicle (PHEV) charging stations in India, using the hybrid Atom Search Woven Aquila Optimization Algorithm (AT-AQ). This method combines ASO with the Aquila Optimizer (AO) to enhance convergence speed and avoid local optima. The AT-AQ algorithm optimizes the placement of charging stations in a renewable energy system, balancing cost and operational efficiency. The model demonstrated superior performance to existing methods, achieving a lower cost function and providing a practical solution for effective charging station distribution.

• Hybridization with Marine Predator Algorithm

Edison Prabhu (2023) introduced a machine-learning model for predicting systemic lupus erythematosus (SLE) symptoms, utilizing IoT data to enhance accuracy. The model integrates a hybrid Marine Predator Algorithm (MPA) and ASO with ANNs. Trained on the Gene Expression Omnibus dataset and validated with patient data, the hybrid approach achieved about 99.70% accuracy, outperforming existing models. This highlights the effectiveness of combining ASO with other optimization techniques for improved disease prediction and management.

• Hybridization with Dolphin Echolocation Algorithm

Deore & Bhosale (2023) proposed the Adaptive DASO to improve intrusion detection systems (IDS). This novel method combines Dolphin Echolocation algorithm with ASO to address the issue of insufficient standard samples. It optimizes feature selection using mutual information and employs a DRNN for intrusion detection. The DRNN, trained with Adaptive DASO, enhances classification accuracy, distinguishing between normal and anomalous network behavior. The approach achieved high performance in specificity, accuracy, and sensitivity, offering a robust solution for IDS.

• Hybridization with Electromagnetic Force Optimization

Dohare & Singh (2020) developed the energy-balanced integrated ASO and EFO (iASEF) scheme to address rapid energy depletion in IoT sensor networks. The approach integrates the ASO algorithm for selecting effective cluster heads and Electromagnetic Force Optimization (EFO) to determine optimal routing paths. The iASEF scheme uses a linear programming model for clustering, considering node degree, intra-cluster distance, residual energy, and inter-cluster distance. Simulation results showed that iASEF outperforms existing algorithms, significantly improving energy efficiency in sensor networks.

• Hybridization with Henry Gas Solubility Optimization

Navin et al. (2023) developed a clinical decision support system for diagnosing congenital heart failure by integrating a robust feature selection mechanism. The authors used a hybrid approach combining HGSO and ASO to enhance diagnostic accuracy. ASO was applied in the wrapper-based feature selection process to optimize classification performance, measured by accuracy and G-mean. This innovative use of ASO highlights its effectiveness in healthcare, improving the precision of machine learning models for early detection of cardiovascular conditions.

• Hybridization with Harris Hawk Optimization

Diab et al. (2020) developed a hybrid control algorithm for Cascaded Modular Multilevel Converters (MMC), integrating HHO and ASO. This approach targets issues such as circulating currents and capacitor stress. The hybrid controller enhances capacitor voltage balancing, reduces harmonic distortion, and minimizes switching frequency. It was tested with an MMC-static synchronous compensator (STATCOM), showing effective stability under different load conditions. The method demonstrates robust performance in controlling MMC systems.

Multi-objective versions of atom search optimization

Previous studies have demonstrated the effectiveness of the ASO algorithm in solving single-objective optimization problems with continuous search spaces. This success has motivated researchers to explore its potential for addressing multi-objective optimization challenges. Junsittiwate, Srinophakun & Sukpancharoen (2022) applied a multi-objective ASO algorithm to optimize palm empty fruit bunch (PEFB) pyrolysis for bio-oil production. Using Aspen Plus simulations and central composite design (CCD), the study evaluated biodiesel yield, CO2 emissions, and utility costs. ASO was employed to maximize yield while minimizing emissions and costs. Compared to CCD, ASO delivered superior results, with a 7.01% increase in biodiesel yield, a 1.33% reduction in CO2 emissions, and a 5.03% reduction in utility costs, showcasing ASO’s efficiency in process optimization.

Asna et al. (2022) introduced a multi-objective binary ASO (MO-QASO), enhanced with quantum operations for optimizing the placement of EV fast-charging stations. The method integrates non-dominated sorting and Pareto optimization to address multiple objectives, such as minimizing grid power loss, station costs, and EV travel time. The quantum-enhanced binarization improves convergence, and the algorithm outperforms traditional binary ASO approaches, demonstrating superior performance in convergence and solution diversity metrics, closely aligning with exhaustive search methods.

Engineering

During its development, Zhao, Wang & Zhang (2019b) tested the performance of ASO by applying it to hydrogeologic parameter estimation and dispersion coefficient estimation (Zhao, Wang & Zhang, 2019a) problems. ASO demonstrated superior performance to other popular algorithms, such as PSO and GA, highlighting its effectiveness and competitiveness in solving real-world optimization problems.

Abde-Rahim, Shaaban & Raglend (2019) utilized ASO to address the optimal power flow (OPF) problem. The OPF-ASO method was tested on the IEEE 30-bus system and showed superior performance to other techniques, underscoring ASO’s robustness and effectiveness in solving complex power flow optimization issues.

Kamel et al. (2019) applied the ASO algorithm to solve the optimal distribution system reconfiguration problem, focusing on minimizing power losses in electrical networks. ASO adjusted the network topology to find configurations that reduce active power losses. The study demonstrated that ASO outperformed other optimization methods, proving its effectiveness in enhancing power distribution systems.

Irudayaraj et al. (2020) employed the ASO algorithm to optimize an FOPID controller for Automatic Load Frequency Control in Hybrid Renewable Energy Systems (HRES). The ASO-tuned FOPID controller improved performance by minimizing the Integral Time Absolute Error, particularly addressing frequency issues arising from intermittent renewable sources and Plug-in Electric Vehicles (EVs). Simulations in MATLAB/Simulink and hardware-in-the-loop tests validated the controller’s enhanced transient and steady-state responses, demonstrating its robustness and real-time applicability.

Khokhar, Dahiya & Parmar (2020) utilized the ASO algorithm for load frequency regulation in HRES by optimizing an integer-order PID controller with ASO. The ASO-based controller exhibited superior performance compared to other approaches in minimizing frequency variations caused by load and wind power fluctuations. The effectiveness of the controller was validated using Bode diagrams and eigenvalue analysis, confirming its robustness in enhancing frequency control for HRES.

Diab et al. (2020) developed a hybrid control algorithm for Cascaded Modular Multilevel Converters (MMC), integrating the HHO and ASO algorithms. This approach addresses issues such as circulating currents and capacitor stress. The hybrid controller enhances capacitor voltage balancing, reduces harmonic distortion, and minimizes switching frequency. It was tested with an MMC-static synchronous compensator (STATCOM), demonstrating adequate stability under different load conditions. The method shows robust performance in controlling MMC systems.

Gour et al. (2020) proposed an optimization framework for improving the placement and sizing of distributed generation (DG) units to minimize power loss in distribution systems. The authors used loss sensitivity factors (LSF) to identify optimal DG locations and applied the ASO algorithm to optimize power loss reduction while adhering to system constraints. Tested on 10-bus and 33-bus radial distribution systems (RDS), the ASO-based framework demonstrated significant improvements in power loss reduction, highlighting ASO’s effectiveness in optimizing complex energy systems.

Deb et al. (2020) proposed using the ASO algorithm to address transmission line congestion in deregulated electricity markets by rescheduling generator outputs. ASO helps optimize real power adjustments to manage congestion effectively. Evaluated on the IEEE 30-bus test system, the method demonstrated improved efficiency and reliability in power transfer, alleviating congestion and enhancing overall system performance.

By integrating the ASO algorithm with ANN, Chaabene & Nehdi (2020) presented a technique for estimating the shear capacity of steel fiber-reinforced concrete (SFRC) beams. ASO optimizes ANN parameters for accurate shear strength predictions. Four classification models (naïve Bayes, SVM, decision tree, KNN) were also used to predict failure modes. ASO-ANN showed the highest accuracy in shear strength prediction, and KNN excelled in failure mode classification. This approach improves SFRC design and retrofitting strategies.

Chandan Kumar Shiva, Kumar & Kumar (2020) designed a dynamic controller for load frequency control (LFC) using a three-degree-of-freedom PID (3DOF-PID) controller optimized with the ASO algorithm. Tested on a four-area power system, the ASO-based 3DOF-PID controller showed superior performance compared to traditional methods, improving settling time, peak variance, and oscillation magnitude. This highlights its effectiveness in enhancing LFC applications.

Khokhar, Dahiya & Parmar (2021) developed an LFC scheme for multi-microgrid systems using electric vehicles (EVs) and a tilt integral derivative (TID) controller optimized by ASO algorithm. This approach improved dynamic performance, reduced oscillations, and enhanced stability under various disturbances, outperforming standard algorithms even with $\pm$30% parameter variations.

Amarendra, Srinivas & Rao (2021) used ASO to optimize the placement of FACTS devices (UPFC, TCSC, IPFC, SVC) to enhance power system security. The ASO-based method was evaluated on IEEE 30, 118, and 300 bus systems using metrics like severity index and voltage stability. ASO demonstrated superior performance in optimizing FACTS device placement compared to conventional algorithms, including DA, WOA, Jaya, FPA, GWO, and JA-FPA.

Kumar & Sinha (2021) proposed a new method for managing power demand and voltage stability in an autonomous microgrid with renewable sources like solar dish-stirling generators. They incorporated battery storage to handle periods of insufficient solar energy and used a classical PI controller for simplicity. The ASO algorithm was employed to optimize power management and voltage stability, showing superior results compared to PSO. Simulations on MATLAB/Simulink validated the effectiveness of the ASO-based approach.

Al-Akayshee, Kuznetsov & Sultan (2021) developed an off-grid hybrid energy system using PV, biomass, and hydroelectric pumped storage to meet power needs in remote Iraq. The author utilized the ASO algorithm to determine the optimal system size and minimize net present cost (NPC), considering real-time load and meteorological data. The ASO-based approach provided an economical and sustainable solution for isolated areas, supporting the development of reliable, clean energy systems in rural regions.

Goud & Rao (2021) introduced an approach to improve power quality (PQ) in HRES using the ASO algorithm and a unified power quality conditioner (UPQC). The authors optimized the UPQC with a FOPID controller designed via ASO for both series and shunt filters. The MATLAB/Simulink simulations showed that the ASO-based method effectively mitigated PQ issues such as voltage sags, swells, and total harmonic distortion (THD), outperforming traditional PI controllers and other optimization methods.

Khonturaev et al. (2021) developed a hybrid approach for optimizing the sizing and placement of distributed generations (DG) in RDS by combining the ASO algorithm with a power loss sensitivity (PLS) index. This approach uses ASO to find optimal DG locations and sizes while the PLS index identifies the best buses for installation. Tested on the IEEE 33-bus RDS, the method improved voltage profiles and reduced power losses significantly, outperforming existing techniques in enhancing grid efficiency and sustainability.

Rao, Srikanth Goud & Rami Reddy (2021) integrated an HRES with grid systems to enhance electricity demand management and reduce environmental impact. The authors used the ASO algorithm to optimize control parameters for a Unified Power Flow Controller (UPQC), addressing issues like voltage sag, swell, and total harmonic distortion (THD). The ASO-optimized fractional-order PID controller outperformed conventional PI controllers, with MATLAB/Simulink simulations confirming its effectiveness in improving power quality in grid-connected HRES.

Reddy et al. (2021) introduced an intelligent control approach for enhancing power quality in grid-connected hybrid power systems with solar, wind, and battery storage. The authors used a Unified Power Quality Conditioner with Active and Reactive Power (UPQC-PQ) combined with an FOPID controller optimized by the ASO algorithm. This method regulates voltage, minimizes power loss, and reduces THD. The ASO-based FOPID controller outperformed traditional controllers and optimization techniques, demonstrating superior performance in various operating modes through MATLAB/Simulink simulations.

Khatir et al. (2021) proposed a damage detection methodology using a flexibility index to locate and quantify damage in complex truss structures. The authors applied SSA and ASO algorithms based on enhanced damage indicators to solve the inverse problem. ASO showed superior convergence performance and lower CPU time compared to SSA. The method was validated on the Guangzhou TV Tower, demonstrating its accuracy and efficiency in predicting damage across different structural floors.

Ranjitha, Ponnurangam & Rajapandiyan (2021) proposed an advanced control strategy for load frequency management in a hybrid power system with wind, solar, and thermal sources. Their method integrates a cascaded PID controller with an ASO-tuned PID controller to improve performance and cost management. The strategy also uses a firefly-based controller for thermal power optimization. This approach demonstrated superior effectiveness to traditional PID and fuzzy logic controllers in simulations, efficiently handling system stability and cost optimization despite renewable energy fluctuations.

Gupta et al. (2021) comprehensively evaluated nine metaheuristic algorithms, including the ASO algorithm, for complex mechanical design problems. The study assessed algorithms such as the Moth-Flame Optimizer (MFO), Multi-Verse Optimizer (MVO), Salp Swarm Algorithm (SSA), and others across eight design challenges. The evaluation focused on solution quality and convergence characteristics, revealing the effectiveness of these algorithms, including ASO, in solving real-world mechanical design issues and highlighting their broad potential for tackling complex optimization tasks.

Duan et al. (2022a) used the ASO algorithm to design gravity-driven drip irrigation systems. The model focuses on minimizing costs while ensuring operational efficiency by optimizing variables such as pipe layout, diameter, emitter-flow rate, and water pressure. Applied to a two-level irrigation project in Xinjiang, ASO effectively reduced the cost per unit area and improved efficiency. The study showed that cost efficiency improved with greater emitter and pipe spacing but decreased with higher emitter-flow rates, highlighting ASO’s effectiveness in optimizing irrigation systems for both cost and performance.

Bahr et al. (2022) proposed an advanced method for managing the dynamic performance and stability of an isolated hybrid micro-grid (HMG) system using solar, wind, and biomass energy sources. They optimized frequency and power control with PID and PIDA controllers, fine-tuning them with the ASO algorithm to minimize the integral of time-weighted absolute error (ITAE). MATLAB/Simulink simulations showed that the ASO-optimized PIDA controller outperformed the conventional PID controller, offering superior frequency stability and reliability under various system uncertainties.

Bagade & Rampure (2022) developed an optimization strategy for enhancing the efficiency of a micro-combined cooling, heating, and power generation (MCCHP) system using the ASO algorithm. The approach optimizes variables in a single-stage, double-effect LiBr-H2O absorption chiller, integrating components such as a diesel engine, heat exchanger, and boiler. The ASO-optimized system achieved a 16 kW cooling capacity with a 98% improvement in temperature reduction and enhanced cost-effectiveness. This optimization significantly benefits energy efficiency and economic viability in hospitals and food production.

Mishra et al. (2022) introduced an ASO-based multistage PID (M-PID) controller for frequency control in multi-microgrid systems with variable renewable energy sources. Their system includes PV panels, wind turbines, diesel generators, and energy storage devices. The ASO-based M-PID controller was compared to GA and PSO, showing superior performance in frequency management. Validation through OPAL-RT simulations demonstrated its effectiveness in handling stochastic conditions and improving control performance.

Tahiliani & Gupta (2022) developed an ASO-based method for optimizing the placement of distributed generators (DG) and distribution static compensators (D-STATCOM) in unbalanced radial distribution systems. Their approach aims to enhance voltage profiles, reduce power losses, and assess cost savings. ASO is used to determine optimal DG and D-STATCOM placement under varying conditions, improving overall power distribution system efficiency. The method’s effectiveness was validated using MATLAB simulations on a 25-bus network.

Bie et al. (2022) introduced a fault diagnosis method for gearboxes that combines the ASO algorithm with resonance-based sparse signal decomposition (RSSD) and support vector machine (SVM) techniques. ASO optimizes the quality factor in RSSD to enhance fault detection by decomposing vibration signals and calculating entropy measures. SVM is then used for fault pattern recognition. The approach demonstrated improved accuracy in identifying gearbox faults in the presence of interference, as shown by simulations and experimental results.

Jiang et al. (2020) developed a hybrid forecasting framework combining Fuzzy Time Series (FTS) with the ASO algorithm to improve forecasting of inbound tourism demand in China, especially with small sample sizes. The framework uses FTS for better information recognition and ASO for optimizing FTS parameters, enhancing forecasting accuracy. Validation through comparative experiments showed that this approach outperforms traditional models, providing reliable forecasts and aiding better planning for the tourism industry.

Dey, Banerjee & Dey (2022) utilized the ASO algorithm to optimize PID controllers for an unstable Magnetic Ball Levitation System (Maglev). The study focused on various PID types, including integer order PID (IOPID), IOPID with a derivative filter (IOPID-F), fractional order PID (FOPID), and FOPID with a fractional order filter (FOPID-F). ASO was employed to fine-tune controller parameters for enhanced stability and transient performance. Results showed that the FOPID-F controller achieved the best stability, input tracking, and robustness, demonstrating ASO’s effectiveness in tackling complex engineering challenges.

Ekinci et al. (2023) proposed using the ASO algorithm for designing digital infinite impulse response (IIR) filters in digital signal processing. The study compared ASO with other metaheuristic algorithms, such as moth flame optimization and gravitational search algorithms. As evidenced by pole-zero diagrams, ASO demonstrated the highest improvement in filter design performance and superior stability. This highlights ASO’s effectiveness and robustness in optimizing IIR filters, outperforming other methods.

Goud et al. (2023) conducted a comprehensive analysis of optimization techniques, including the ASO algorithm, to enhance power quality in electrical systems. The study reviews various metaheuristics like Whale Optimization Algorithm (WOA), Black Widow Optimization (BWO), Gravitation Search Algorithm (GSA), and ASO, emphasizing their effectiveness in managing complex nonlinear systems. The article demonstrates the significant potential of these algorithms, especially in optimizing power quality when integrating renewable energy and storage systems into power networks.

Kala et al. (2023) proposed an ASO-based method for selective harmonic elimination (SHE) in multilevel inverters (MLI). This approach uses ASO to optimize firing angles in an 11-level MLI system, effectively reducing low-order harmonics while maintaining fundamental voltage control. Compared to other metaheuristic techniques such as bee algorithm, ICA, firefly optimizer, PSO, and TLBO, the ASO method demonstrated superior performance in reducing total harmonic distortion (THD) and harmonics, with successful results from simulations and experiments on a photovoltaic-based RCC MLI.

Isabella, Umamaheswari & Marimuthu (2023) introduced a high-gain BOOST-CUK (BOCUK) DC-DC converter for switched mode power supply (SMPS) applications, combining Boost and Cuk converters for better voltage gain and efficiency. The authors optimized the performance using an ASO-tuned PI controller within a dual-loop control system. This approach improved line and load fluctuations regulation, demonstrating ASO’s potential in refining control systems for efficient and stable power conversion with reduced harmonic distortion and improved power factor.

Pavani et al. (2023) applied the ASO algorithm to synthesize linear and planar antenna arrays, optimizing symmetric and asymmetric linear arrays and two-ring concentric circular arrays with non-uniform amplitude coefficients. ASO outperformed other algorithms, such as the Flower Pollination Algorithm (FPA) and Firefly Algorithm (FA) in pattern synthesis. This demonstrates ASO’s effectiveness in solving complex electromagnetic optimization problems, making it a valuable tool in antenna design.

Sepperumal et al. (2023) proposed an ASO-tuned FOPI controller for enhancing a DC–DC SEPIC converter used in power factor correction and voltage regulation. The ASO algorithm optimized the controller’s parameters in voltage and current control loops, resulting in superior dynamic response, efficiency, and power factor compared to a PSO-optimized FOPI controller. The ASO-tuned controller also demonstrated improved stability and voltage regulation under varying loads, showcasing ASO’s effectiveness in optimizing power electronics control systems.

Chen et al. (2023b) proposed an ASO-Back Propagation (ASO-BP) neural network to predict wear rates of copper-based powder metallurgy brake pads. By optimizing the weights and thresholds of a Back Propagation neural network with ASO, the model achieved 97.3% prediction accuracy. The study highlighted the significant impact of braking speed and pressure on performance, improving the reliability of brake systems in high-speed trains.

Sekar et al. (2023) developed an advanced method to enhance power quality and reactive power compensation in grid-connected PV systems. The authors used the ASO algorithm for MPPT to optimize energy extraction from PV panels. The approach includes a non-isolated high voltage gain DC-DC converter and a Coyote Optimized Converter Control (COCC) for precise switching. A residual attention echo state reactive controller (RaERC) also improves control signals for a nine-level inverter. Simulations validated significant improvements in power quality and efficiency.

Wang et al. (2023) introduced a planetary gear fault diagnosis method for wind turbines using a digital twin model integrated with ASO algorithm and SVM techniques. The system combines real-time and virtual data to improve fault detection, overcoming the limitations of traditional methods. Using empirical mode decomposition (EMD) with ASO-SVM, the approach enhances diagnostic accuracy and operational efficiency. Implemented on Unity3D, it offers advanced real-time monitoring and early-warning capabilities for wind turbine maintenance.

Osama Abed El-Raouf et al. (2023) proposed enhancing grid-tied renewable energy systems with a unified power flow controller (UPFC) regulated by a FOPID controller. The authors optimized the FOPID parameters using the ASO algorithm. The study showed that the ASO-tuned FOPID controller improved system stability and power quality during disturbances, reduced total harmonic distortion, and supported maximum power point tracking for photovoltaic cells and wind turbines.

Kumar & Sikander (2023) introduced a method for creating reduced-order models (ROMs) of large-scale linear time-invariant (LTI) systems using the ASO algorithm. The ASO technique simplifies single-input single-output (SISO) systems by minimizing the integral square error (ISE) between the original and reduced-order models. Using molecular dynamics to optimize search variables, the ASO-based ROMs demonstrated superior time and frequency response compared to traditional methods. This approach highlights ASO’s efficiency and accuracy in control engineering applications.

Hao et al. (2023) proposed an optimization model for short-term traffic flow prediction, integrating the ASO algorithm with Extreme Learning Machine (ELM). The ASO-ELM model optimizes ELM performance, addressing the challenges of dynamic and nonlinear traffic flow. Tested with A10 ring road data in Amsterdam, the model improved prediction accuracy, reducing mean absolute percentage error (MAPE) by 9.8% and root mean squared error (RMSE) by 5.9% compared to traditional models like ANN. This approach highlights the effectiveness of combining ASO with ELM for accurate traffic flow forecasting.

Saadi et al. (2023) developed a high-performance control strategy for an interleaved non-isolated DC/DC converter in fuel cell applications, using the ASO algorithm for controller design. The strategy features a two-degree-of-freedom PID outer loop for setpoint tracking and disturbance rejection and a super-twisting integral sliding mode (STISM) inner loop for rapid convergence and reduced chattering. ASO optimizes controller gains for robust and stable operation. Validated through simulations and real-time tests, the ASO-based approach outperforms other methods and dual-loop PID controllers in performance.

Zhang et al. (2023a) introduced SPAA, a sample-wise probabilistic approach combining ASO with ANN to assess the stability of circular shafts in urban infrastructure, considering soil variability. This ASO-ANN model replaces extensive simulations with Latin hypercube sampling and iterative enrichment. Validated by Monte-Carlo simulations and global sensitivity analysis, SPAA offers reliable failure probabilities and sensitivity indices while reducing simulation requirements, outperforming traditional methods in high-dimensional scenarios.

Güven & Mengi (2023) developed a framework using the ASO algorithm to optimize HRES in Turkey. The authors integrated wind, solar, biomass, and fuel cells with excess energy stored as hydrogen. ASO was used to fine-tune variables like wind and solar power and hydrogen storage, achieving the most cost-effective solution and best convergence among the ten algorithms. The approach was validated through MATLAB simulations, demonstrating significant improvements in energy flow management and cost reduction.

Mohamed, Hasanien & Ebrahim (2024) improved PV systems’ performance under partial shading by using the ASO algorithm to optimize global MPPT controllers. The authors integrated ASO with Proportional-Integral, Fractional-Order PI, and Fuzzy Logic Controllers, achieving better tracking accuracy, faster convergence, and higher power efficiency. ASO-based Fuzzy IC MPPT controllers showed reduced settling times and improved stability.

Healthcare and medical applications

Surenthiran, Rajalakshmi & Sujatha (2021) introduced a DBNN optimized with the ASO algorithm for predicting student performance. The DBNN uses a cognitive divergence algorithm with restricted Boltzmann machines (RBMs) to enhance learning rate optimization. The model achieved 90% accuracy and reduced error rates to below 20%. It outperformed previous methods, offering valuable insights for early academic interventions.

Marzouk et al. (2022) developed the ASODCAE-SLR model, which combines the ASO algorithm with a deep convolutional autoencoder (DCAE) for Arabic Sign Language (ASL) recognition. The model addresses the challenge of recognizing complex hand gestures using weighted average filtering for input frame preprocessing and a capsule network (CapsNet) for feature extraction. ASO optimizes hyperparameters to enhance DCAE performance. Tested on the Arabic Sign Language dataset, the ASODCAE-SLR model achieved superior recognition accuracy compared to existing methods, improving communication for individuals with hearing and speech disabilities.

Toğaçar (2022) developed an advanced method for detecting diabetic retinopathy by integrating the ASO algorithm with deep learning techniques. Their approach involves preprocessing fundus images using morphological gradients and segmentation to enhance image quality and extract ocular vessel information. Transfer learning is employed to train CNNs on publicly available datasets, extracting activation sets for diabetic retinopathy severity classes. ASO is then used to select the most influential activation sets, improving classification performance. This method achieved accuracy rates of 99.59% and 99.81% on two separate datasets, significantly enhancing the early detection of diabetic retinopathy.

Vadlamudi et al. (2023) proposed an advanced heart disease detection model using deep learning techniques for ECG signal analysis. The model includes pre-processing with a discrete wavelet transform (DWT) filter, feature extraction through CNN, and classification of heart conditions. Although traditional optimization methods like ASO and GWO were used, the deep learning approach demonstrated superior performance, achieving high sensitivity, accuracy, and recall. This approach proves to be more competitive in heart disease prediction than existing techniques.

Ali et al. (2024) proposed a wrapper-based technique to enhance chest infection detection using X-ray images. The authors extracted deep features with pre-trained models and optimized these features using ten algorithms, including ASO, PFA, HGSO, HHO, MRFO, EO, SMA, GNDO, and MPA. Combined with network selection, the method achieved a high classification accuracy of 97.7% on a chest infection X-ray dataset, significantly improving detection performance and demonstrating strong potential for clinical use.

Internet of Things and communication

Almagboul et al. (2019) developed an ASO-based method for optimizing hybrid analog-digital (HAD) beamforming in 5G cellular networks. This approach uses a partially connected HAD beamformer to address challenges such as high energy consumption and hardware costs. ASO optimizes beamforming weights by adjusting digital amplitudes and analog phase shifts to minimize peak sidelobe levels, accurately steer nulls, and reduce power consumption. Simulation results showed that ASO consistently outperformed other metaheuristic algorithms in sidelobe reduction, null depth, and convergence, demonstrating its effectiveness for large-scale antenna systems in 5G networks.

Nath, Som & Negi (2020) introduced an approach to enhance IoT security by combining the ASO algorithm with Elliptic Curve Cryptography (ECC). The ASO-based ECC optimizes key generation to improve encryption and decryption efficiency. This method addresses security vulnerabilities by selecting optimal keys, leading to a reduced encryption time of 2.7384 s and better key management. Simulation results showed that the ASO-enhanced ECC outperformed traditional methods in encryption time, entropy, and Peak Signal-to-Noise Ratio (PSNR), demonstrating superior system security and efficiency.

Elkamchouchi et al. (2022) proposed the ASO-based Clustering with Encryption Technique for Secure Internet of Drones (ASOCE-SIoD) to enhance security and efficiency in Internet of Drones (IoD) environments. ASO is used to form clusters and designate cluster heads, while an encryption technique encrypts images captured by drones. This approach ensures secure transmission of visual data to ground stations. Experimental results showed that ASOCE-SIoD outperformed existing methods, particularly in clustering efficiency and data security within dynamic IoD environments.

Korkmaz & Akgüngör (2022) introduced a new approach for predicting traffic signal cycle lengths, improving the traditional Webster model, which struggles in high-traffic scenarios. The authors developed predictive models for different traffic conditions using ASO and GOA. Evaluated in exponential, power, and quadratic forms, the ASO-based models outperformed the Webster model and VISTRO optimization, providing more accurate predictions and reducing delays. Statistical analyses confirmed that ASO was superior to GOA in predicting optimal cycle lengths.

Al Duhayyim et al. (2022) developed a smart water quality prediction approach using the ASO algorithm with a fuzzy deep convolutional network (F-DCN) in an IoT environment. By optimizing hyperparameters with ASO, their method improved prediction accuracy for water quality parameters monitored by IoT devices. Simulations showed that the ASO-enhanced F-DCN model significantly outperformed existing techniques, offering a promising solution for water quality monitoring in IoT-based systems.

Srinivasiah, Ranganathasharma & Ramanna (2023) introduced the trust-aware clustering and routing protocol (TCRP) for wireless sensor networks (WSNs), leveraging the ASO algorithm to improve data security and energy efficiency. TCRP enhances node security by selecting secure cluster heads (CHs) and establishing protected routing paths with node authentication. The fitness function considers distance, trust, and energy. TCRP outperformed existing protocols such as SQEER and PFCD-ARP-CH, achieving 99% throughput with 80 nodes and demonstrating superior performance in energy consumption, throughput, and packet delivery ratio.

Clustering and data mining

Karadeniz, Çelik & Başaran (2023) developed a method for distinguishing walnut varieties using image processing and optimization. The authors created a dataset of 1,751 leaf images from 18 walnut species and used a residual block-based convolutional neural network to extract deep features. The ASO algorithm was applied to select the most distinguishing features, which were then classified using a Support Vector Machine (SVM), achieving 87.42% accuracy. This approach highlights the effectiveness of combining ASO with deep learning for precise classification in botanical studies.

Mao et al. (2023) introduced the Fast Mahalanobis Classification System (FMCS) for high-dimensional and imbalanced data in hydraulic system fault diagnosis. FMCS employs a two-stage dimensionality reduction method, combining symmetrical uncertainty with Mahalanobis kernel PCA. The Mahalanobis distance assesses feature importance, and the ASO algorithm improves threshold determination in the Mahalanobis-Taguchi system. The FMCS model outperformed 24 baseline methods, achieving high classification accuracy (0.92), an F1 score (0.94), and a G-mean (0.91), with reduced computation time (4.9045 s), demonstrating its effectiveness in intelligent fault diagnosis for hydraulic systems.

Özbay, Özbay & Gharehchopogh (2024) developed a methodology for mushroom species classification using a dataset of 6,714 images from nine species. The authors applied visualization techniques such as Grad-CAM, LIME, and Heatmap to enhance feature extraction. The features were optimized using the ASO algorithm, significantly reducing the feature map size. Coupled with a residual block-based CNN, the ASO-optimized approach achieved a 95.45% classification accuracy with the KNN classifier. This demonstrates ASO’s effectiveness in feature optimization and highlights the benefit of advanced visualization techniques for mushroom species identification.

Environmental modeling

Cao et al. (2022) developed a short-term PV power forecasting model that combines the ASO algorithm with a backpropagation neural network (BPNN). ASO optimizes the BPNN’s weights and thresholds, while the Pearson correlation coefficient is used to select key meteorological factors, such as irradiance and temperature. The Euclidean distance formula customizes training sets for each instance. Applied to a solar farm in Yunnan, China, the ASO-BPNN model demonstrated improved accuracy in forecasting PV power output, enhancing the operational efficiency and stability of grid-connected PV systems.

A hybrid deep learning model called CSVR was presented by Ghimire et al. (2022), which combines NN and SVR to forecast global solar radiation (GSR). CNN extracts features from time series data, while SVR forecasts daily GSR. The ASO algorithm was used to select key meteorological variables, and HyperOpt optimized the hyperparameters. The CSVR model demonstrated superior predictive accuracy with lower errors than other methods, highlighting ASO’s role in improving feature selection and enhancing renewable energy forecasting.

Duan et al. (2022b) developed an optimization model to minimize total investment in a multilevel gravity drip irrigation system’s pipe network, using the ASO algorithm to enhance pipe layout and diameter selection. The model optimizes the field drip irrigation pipe network (FDIPNS) and then refines the main water distribution system (MWDPNS) based on the FDIPNS results. By addressing topological and hydraulic constraints, the ASO-based method improved the pipe network design in Xinjiang, China, demonstrating its practical benefits for managing irrigation systems in irregularly shaped fields.

Kosasaeng et al. (2022) utilized the ASO algorithm, wind-driven optimization, and GP to optimize reservoir rule curves for a multi-reservoir system in Sakon Nakhon, Thailand, addressing the severe flooding of 2017. Their method aimed to reduce excess water release by optimizing reservoir operations using hydrological data and water demand. The application of ASO demonstrated that managing multiple reservoirs was more effective than single-reservoir systems, enhancing water management efficiency without needing new infrastructure.

Ann et al. (2022) developed a hybrid optimization method combining multi-gene symbolic regression with the ASO algorithm to minimize evapotranspiration (ET) in aquaponic lettuce. By optimizing artificial light properties, the approach improved plant growth and water efficiency in controlled-environment agriculture (CEA). ASO, when used with genetic programming, found the optimal solution, enhancing yield and growth. This highlights ASO’s effectiveness in agricultural optimization and resource management in CEA systems.

Li et al. (2023) introduced a PV reconfiguration method using the ASO algorithm to address hot spot effects from partial shading. Evaluated by mismatch loss, fill factor and standard deviation, their method showed better efficiency, speed, and reliability than other techniques. This demonstrates ASO’s potential for optimizing energy systems and enhancing PV array performance and durability under diverse conditions.

Azma et al. (2023) proposed two hybrid models—BBO-ANN and ASO-ANN—to predict dissolved oxygen (DO) levels in the water. Combining ASO with ANN and BBO, these models used 5 years of data from Rock Creek. The ASO-ANN model achieved a mean absolute percentage error (MAPE) of 2.5170% and a correlation of 0.99135, while the BBO-ANN model had a slightly better MAPE of 2.3848% and a correlation of 0.99186. Both models outperformed existing methods, with the BBO-ANN providing a practical formula for DO prediction.

Ghosh et al. (2024) developed a predictive framework for modeling clean electricity generation in India, focusing on hydro, nuclear, and renewable sources. The authors combined ensemble empirical mode decomposition, density-based clustering, and the ASO algorithm to forecast energy production. The framework effectively modeled hydroelectric generation, particularly during volatile periods like the COVID-19 pandemic, highlighting the role of industrial demand in clean energy growth.

Security

Saba et al. (2021) developed an automated exam proctoring system by integrating the ASO algorithm with a deep learning model called L2-GraftNet, which combines AlexNet and SqueezeNet features. ASO optimized feature extraction, and the fine KNN classifier achieved 93.88% accuracy, demonstrating the system’s effectiveness in automating exam monitoring and reducing manual supervision.

Lavanya & Kavitha (2021) introduced an advanced iris recognition framework to improve biometric identification accuracy. The authors used a combination of SUSAN, GHT, and VJ techniques for eye detection, along with fuzzy Retinex for image enhancement. ASO was applied to optimize the matching and classification process in an FFCNN, achieving a high recognition rate of 99.9%, surpassing other methods.

Silivery, Rao & Kumar (2023) developed an intrusion detection system for DoS and DDoS attacks, using deep convolutional generative adversarial networks (DCGAN) to address class imbalance and ResNet-50 for feature extraction. The authors applied ASO to optimize an AlexNet-based model, achieving high accuracy on the CICIDS2019 and UNSW-NB15 datasets, highlighting ASO’s effectiveness in optimizing deep learning models for cybersecurity.