Recent advances in use of bio-inspired jellyfish search algorithm for solving optimization problems

Giuseppe Meazza

This section focuses on the use of jellyfish search optimizer (JSO) in engineering optimization, prediction and classification, and the algorithmic fine-tuning of artificial intelligence.

Engineering optimization

The jellyfish search optimizer (JSO) has been used in various engineering fields, such as power systems and energy generation, communication and networking, and civil and construction engineering as detailed below.

Power system and energy generation

This section examines the uses of JSO in power systems and energy-related fields. Rai and Verma63 used JSO to find the economic load dispatch of generating units, considering transmission losses. The effectiveness of the proposed method was evaluated by testing it on six systems under various loads. It was compared with the lambda-iterative and PSO algorithms to find the most efficient among them. JSO yielded the lowest fuel cost and transmission losses of the compared methods.

Tiwari et al.64 used JSO to analyze the effect of the installation of distributed generation (DG) and a capacitor bank (CB) on a radial distribution system (RDS). They carried out a cost-based analysis that considered the major expenses that are incurred due to the installation, operation, and maintenance of DG and CB units. They tested JSO on the IEEE 33-bus RDS. The results of their simulation were compared with those of simulations of other methods in the literature. JSO outperformed these other methods in terms of both power loss minimization and profit maximization.

Farhat et al.65 used JSO to propose a power flow model that included three types of energy sources, which were thermal power generators representing conventional energy sources, wind power generators (WPGs), and solar photovoltaic generators (SPGs). They used a modified IEEE 30-bus test system to determine its feasibility. To examine the effectiveness of the proposed power flow model, the results of its simulation were compared with the results of simulations of four other nature-inspired global optimization algorithms. The results established the effectiveness of JSO in solving the optimal power flow (OPF) problem with respect to both minimization of total generation cost and solution convergence.

Shaheen et al.66 introduced an efficient and robust technique that used JSO for optimal Volt/VAr coordination based on a joint distribution system reconfiguration (DSR)with the integration of distributed generation units (DGs) and the operation of distribution static VAr compensators (SVCs). JSO yielded the best solution and a comparison of the proposed JSO with similar approaches demonstrated its usefulness in modern control centers.

Alam et al.67 used JSO to track the global maximum power point (GMPP) of the solar photovoltaic (PV) module under partial shading conditions. Their results suggested that the JSO method has good tracking speed and accuracy. They also found that the JSO strategy tracks the GMPP in half of the time that is taken by the PSO algorithm under both uniform and shaded conditions.

Boutasseta et al.68 used the JSO technique to modify the voltage of a photovoltaic array using a boost direct current to direct current (DC–DC) converter. Their experimental results revealed that JSO performs well under both normal and disturbed operating conditions.

Abdulnasser et al.69 used JSO for the optimal sizing and placement of DGs and capacitor banks (CBs). To elucidate the efficiency of the proposed algorithm, they considered various cases; the allocation of CBs only, the allocation of DGs only, and the allocation of both CBs and DGs. The results thus obtained reveal that the JSO delivered the best results with respect to technical, economic, and emission objectives.

Ngo70 used JSO to solve the economic dispatch problem and to reduce costs and fuel consumption in power systems. To verify the feasibility and the effectiveness of the proposed scheme, they conducted two case studies to test the optimization performance of the proposed method from multiple economic perspectives. The validation results reveal that the proposed scheme provided more speed, convergence, and robustness than the methods to which it was compared.

Nusair et al.71 used JSO and other recently developed algorithms (slime mould algorithm (SMA), artificial ecosystem-based optimization (AEO), and marine predators algorithm (MPA)) to solve both multi- and single-OPF objective problems for a power network that incorporate flexible alternating current transmission system (FACTS) and stochastic renewable energy sources. They compared these algorithms to commonly available alternatives in the literature such as PSO, moth flame optimization (MFO), and grey wolf optimization (GWO), using an IEEE 30-bus test system. Their results reveal that JSO and other recently presented algorithms (MPA, SMA, and AEO) are more effective than PSO, GWO, and MFO in solving OPF problems.

Eid72 used JSO to allocate distributed generators (DG) and shunt capacitor (SC) banks optimally in distribution systems. He found JSO to be practical and effective in solving such nonlinear optimization problems, yielding better results than other algorithms in the literature. Huang and Lin73 used an improved jellyfish search optimizer (IJSO) to track the maximum power point (MPPT) under partial shade conditions. Their results showed that IJSO can accurately track the global maximum power point, and that it converged more quickly than an improved particle swarm optimization (IPSO) algorithm.

Shaheen et al.57 integrated a novel amalgamated heap-based agent with jellyfish optimizer (AHJFO) to optimize combined heat and power economic dispatch (CHPED). They improved the efficiency of two newly developed techniques: the heap-based optimizer (HBO) and JSO. AHJFO incorporates an adjustment strategy function (ASF) to improve exploration in a few iterations by improving the solutions that are generated using HBO. As the iterations proceed, exploitation is improved by updating solutions that are generated using JSO. AHJFO is more effective than HBO and JSO in solving the CHPED problem for medium-sized 24-unit and large 96-unit systems. Simulation results reveal the superiority of the proposed AHJFO over HBO, JSO and other algorithms for solving the CHPED problems.

Ginidi et al.58 proposed an innovative hybrid heap-based JSO (HBJSO) to improve upon the performance of two recently developed algorithms: the heap-based optimizer (HBO) and JSO. HBJSO uses the explorative features of HBO and the exploitative features of JSO to overcome some of the weaknesses of these algorithms in their standard forms. HBJSO, HBO, and JSO were validated and statistically compared by using them to solve a real-world optimization problem of combined heat and power (CHP) economic dispatch. HBJSO, HBO, and JSO were applied to two medium-sized 24-unit and 48-unit systems, and two large 84 unit and 96-unit systems. The experimental results demonstrate that the proposed hybrid HBJSO outperforms the standard HBO, JSO and other reported techniques when, applied to CHP economic dispatch.

Shaheen et al.74 proposed an enhanced quasi-reflection jellyfish optimization (QRJFO) algorithm for solving the optimal power flow problem. Fuel costs, transmission losses and pollutant emissions were considered as multi-objective functions. The performance of the proposed QRJFO algorithm was evaluated on the IEEE 57-bus, the practical West Delta Region system and a large IEEE 118-bus. Simulation results demonstrate the quality of the solutions and resilience of QRJFO.

Boriratrit et al.75 used jellyfish search extreme learning machine (JS-ELM), the Harris hawk extreme learning machine (HH-ELM), and the flower pollination extreme learning machine (FP-ELM) to increase accuracy and reduce overfitting in electric energy demand forecasting. Their results show that the JS-ELM provided a better minimum root mean square error than the state-of-the-art forecasting models.

Ali et al.76 presented an effective optimal sizing technique for a hybrid micro-grid using JSO. Their proposed sizing approach considers uncertainty associated with hybrid renewable resources. They investigated several operating scenarios to evaluate the effectiveness of the proposed approach and compared it to various optimization techniques. Their results demonstrate the applicability of JSO to their problem of interest.

Rai and Verma77 used JSO to solve a combined economic emission problem for an isolated micro-grid. They conducted tests on this micro-grid system, comprising traditional generators and renewable energy sources in two scenarios. They compared the results with those obtained using available algorithms to prove that the JSO algorithm was more effective than the others.

Yuan et al.78 used the improved jellyfish search optimizer and support vector regression (IJSO-SVR) to solve the problems of grid connection and power dispatching that are caused by non stationary wind power output. IJSO exhibits good convergence ability, search stability, and optimum-seeking ability, and it is more effective than conventional methods in solving optimization problems. The IJSO-SVR model outperformed other models in the literature and presents a more economical and effective means of optimizing wind power generation to solve problems with its uncertainty and can be used in grid power generation planning and power system economic dispatch.

Chou et al.79 used JSO and convolutional neural networks (CNNs) to evaluate the power generation capacity of plant microbial fuel cells (PMFCs) on building rooftops. Their results demonstrate the superior performance of JSO-optimized deep CNNs in learning image features and their consequent suitability for constructing models for estimating power generation by PMFCs.

Communication and networking

This section investigates the use of JSO in the field of communication and networking. Selvakumar and Manivannan80 used JSO to overcome the shortcomings of defragmentation in networking, and to improve the quality of network services. The proposed combination of proactive/reactive defragmentation approach and JSO (PR-DF-JSO) outperformed state-of-the-art spectrum defragmentation algorithms in terms of spectrum utilization, network efficiency, and quality of service offered based on the results of experiments and standard quality metrics. Specifically, lower spectrum fragmentation complexity, a better bandwidth fragmentation ratio, and less overall connection blocking were achieved.

Durmus et al.81 used swarm-based metaheuristic algorithms JSO, PSO, artificial bee colony (ABC), and the mayfly algorithm (MA), to determine the optimal design of linear antenna arrays. They conducted extensive experiments on the design of 10-, 16-, 24- and 32-element linear arrays and determined the amplitude and the positions of the antennas. They performed each of their experiments 30 times owing to the randomness of swarm-based optimizers, and their statistical results revealed that the novel algorithms JSO and MA outperformed the well-known PSO and ABC methods.

Aravind and Maddikunta82 proposed a novel optimal route selection model for use with the internet of things (IoT) in the field of healthcare that was based on an optimized adaptive neuro-fuzzy inference system (ANFIS). They selected optimal routes for medical data using a new self-adaptive jellyfish search optimizer (SA-JSO) that was an enhanced version of the original JSO algorithm39. Their model outperformed others.

Civil and construction engineering

Structural optimization has become one of the most important and challenging branches of structural engineering, and it has consequently received considerable attention in the last few decades83. Chou and Truong39 developed JSO, motivated by the behavior of jellyfish in the ocean for use in civil and construction engineering. They used JSO to solve structural optimization problems, including 25, 52, and 582-bar tower designs. Their results showed that JSO not only performed best but also required the fewest evaluations of objective functions. Therefore, JSO is potentially an excellent metaheuristic algorithm for solving structural optimization problems.

Chou and Truong46 expanded the framework of the single-objective jellyfish search (SOJS) algorithm to a multi-objective jellyfish search optimizer (MOJS) for solving engineering problems with multiple objectives. MOJS integrates Lévy flight, an elite population, a fixed-size archive, a chaotic map, and the opposition-based jumping method to obtain Pareto-optimal solutions. Three constrained structural problems (25, 160, and 942-bar tower designs) of minimizing structural weight and maximum nodal deflection have been solved using the MOJS algorithm. MOJS is an effective and efficient algorithm for solving multi-objective optimization problems in civil and construction engineering.

Kaveh et al.52 proposed a quantum-based JSO, named Quantum JSO (QJSO), for solving structural optimization problems. QJSO is used to solve frequency-constrained large cyclic symmetric dome optimization problems. The results thus obtained reveal that QJSO outperforms the original JSO and has superior or comparable performance to that of other state-of-the-art optimization algorithms.

Rajpurohit and Sharma56 proposed an enhancement of JSO by the implementation of chaotic maps in population initialization. They applied their enhanced JSO to three classical constrained engineering design problems. Analysis of the results suggests that the sinusoidal map outperforms other chaotic maps in JSO and helps to find efficiently the minimum weight design of a transmission tower.

Ezzeldin et al.84 used JSO to develop optimal strategies for the sustainable management of saltwater intrusion into coastal aquifers based on the finite element method (FEM). They tested the effectiveness of JSO by applying it to a real aquifer system in Miami Beach to maximize its total economic benefit and total pumping rate. JSO has also been used in a case study of the El-Arish Rafah aquifer, Egypt, to maximize the total pumping rate. The results in both cases were compared to relevant results in the literature, revealing that the JSO is an effective and efficient management tool.

Chou et al.85 used JSO and convolutional neural networks (CNNs) to predict the compressive strength of ready-mixed concrete. Their analytical results reveal that computer vision-based CNNs outperform numerical data-based deep neural networks (DNNs). Thus, the bio-inspired optimization of computer vision-based convolutional neural networks has promise for predicting the compressive strength of ready-mixed concrete.

Chou et al.86 presented jellyfish search optimizer (JSO)-XGBoost and symbiotic organisms search (SOS)-XGBoost for forecasting the nominal shear capacity of reinforced concrete walls in buildings. Their proposed methods outperform the ACI provision equation and grid search optimization (GSO)-XGBoost in the literature. Thus, they can be used to improve building safety, simplify a cumbersome shear capacity calculation process, and reduce material costs. Their systematic approach also provides a general framework for quantifying the performance of various mechanical models and empirical formulas that are used in design standards.

Truong and Chou47 proposed a novel fuzzy adaptive jellyfish search optimizer (FAJSO) for use in the stacking system (SS) of machine learning. They integrated the JSO, the fuzzy adaptive (FA) logic controller, and stacking ensemble machine learning. Its application to construction productivity, the compressive strength of a masonry structure, the shear capacity of reinforced deep beams, the axial strength of steel tube confined concrete, and the resilient modulus of subgrade soils was investigated. Their results indicate that the FAJSO-SS outperformed other methods. Accordingly, their proposed fuzzy adaptive metaheuristic optimized stacking system is effective for providing engineering informatics in the planning and design phase.

Prediction and classification

Prediction and classification are required in a variety of areas that involve time series and cross-sectional data87, 88. This section concerns articles in which JSO has been used alone or integrated with machine/deep learning algorithms for prediction and classification.

Almodfer et al.89 employed a random vector functional link (RVFL) network that was optimized by JSO, AEO, MRFO, and SCA to predict the performance of a solar thermo-electric air-conditioning system (STEACS). Their results revealed that the RVFL-JSO outperformed the other algorithms in predicting all responses of the STEACS with a correlation coefficient of 0.948–0.999. They recommended its use for modeling STEACS systems.

Chou et al.90 used JSO to optimize the convolutional neural network (CNN) hyper-parameters to ensure the accuracy and stability of CNN in predicting power consumption. Their analytical results provide insights into the formulation of energy policy for management units and can help power supply agencies to distribute regional power in a way that minimizes unnecessary energy loss.

Barshandeh et al.91 utilized JSO and the marine predator algorithm (MPA) to develop a learning-automata (LA)-based hybrid algorithm for benchmark function optimization and solving data clustering problem. They applied the proposed algorithm to ten datasets and compared it with competing algorithms using various metrics; the hybrid algorithm outperformed. Desuky et al.92 used JSO to classify imbalanced and balanced datasets. They performed experiments on 18 real imbalanced datasets, and the proposed method performed comparably with well-known and recently developed techniques.

Chou and Truong88 tested JSO and other parameter-less algorithms (TLBO, SOS) by using them in the hyperparameters finetuning of least squares support vector regression (LSSVR) to develop a novel forecasting system. The linear time-series has been optimized using nonlinear machine learning models to identify historical patterns of regional energy consumption. Analytical results confirm that the proposed system, JSO-LSSVR, can predict multi-step-ahead energy consumption time series more accurately than can the linear model.

Chou et al.93 developed a weighted-feature least squares support vector regression (WFLSSVR) model that is optimized by JSO to predict the peak friction angle (shear strength) of fiber-reinforced soil (FRS), which is a popular material for use in building geotechnical structures. Their results showed that JSO-WFLSSVR outperformed baseline, ensemble, and hybrid machine learning models, as well as empirical methods in the literature. The JSO-WFLSSVR model is also effective for selecting features and can help geotechnical engineers to estimate the shear strength of FRS.

Hoang et al.94 implemented a support vector machine classifier that was optimized using JSO for the automatic classification of the severity of concrete spalling. It partitions input data into two classes, shallow spalling and deep spalling. Experimental results, supported by the Wilcoxon signed-rank test, reveal that the newly developed method is highly effective for classifying the severity of concrete spalling with an accuracy rate of 93.33%, an F1 score of 0.93, and an area under the receiver operating characteristic curve of 0.97.

Siddiqui et al.95 used JSO to calculate the optimum switching angle in the modulation range to eliminate desired lower-order harmonics in a multilevel inverter (MLI) voltage control application. The total harmonic distortion (THD) values of five-, seven-, and nine-level were computed using JSO and compared with those obtained using the powerful differential evolution (DE) algorithm. The results thus obtained clearly demonstrated that the output of an MLI in JSO exhibits THD that is superior to that in the output of DE for low and medium values of the modulation index.

Çetinkaya and Duran96 used JSO and other recently developed optimization algorithms [marine predators’ algorithm (MPA), tunicate swarm algorithm (TSA), mayfly optimization algorithm (MOA), chimp optimization algorithm (COA), slime mould optimization algorithm (SMOA), archimedes optimization algorithm (AOA), and equilibrium optimizer algorithm (EOA)] to improve the precision of the clustering-based segmentation of vessels. Simulation results of these algorithms exhibited similar convergence rates and error performances. Statistical analyses demonstrated that the stability and robustness of each metaheuristic approach sufficed to separate vessel pixels from the background pixels of a retinal image.

Wang and Gao97 used the multi-objective jellyfish search optimizer (MOJS) to determine the weights of kernel functions. According to their experimental result concerning three American solar sites, the proposed system that integrates with MOJS provided a higher interval coverage rate and a narrower interval width than those of other systems.

Zhao98 used single-objective JSO to classify brain function in human brain function parcellation. Experimental results show that that the new method not only has a greater searching ability than other partitioning methods, but also can obtain better spatial structures and stronger functional consistency.

Lei et al.61 proposed an enhanced algorithm, known as the fractional-order modified strategy and Gaussian mutation mechanism jellyfish search optimizer (FOGJSO), to predict rural resident income. They used FOGJSO to optimize the order of a discrete fractional time-delayed grey model for forecasting rural resident income. The results reveal that FOGJSO performed much better with respect to precision and convergence speed than did other methods.

Shubham et al.99 used JSO for clustering between a dish type stirling solar generator, a micro hydro turbine, a diesel generator, a flywheel energy storage device, a super magnetic energy storage device and an electric vehicle in a renewable energy based microgrid to stabilize the frequency and tie line power in the system. They compared the performance of the JSO based dual stage controller with those of the black widow optimization algorithm, GA and the PSO-based controller, with respect to overshoot, undershoot, settling time and figure of demerit. JSO outperformed other optimization algorithms when used to tune dual stage (1+PI)TID controller involving a microgrid-based electric vehicle.

Finetuning of artificial intelligence

Hyper-parameter optimization is essential to the development of efficient models in machine learning and deep learning algorithms, as well as for quality control in industrial production100, 101. JSO is an efficient and innovative algorithm that is used in hyper-parameter optimization.

Chou et al.102 used JSO to optimize the hyper-parameters of a deep learning model that is called residual network (ResNet) and is used to classify the deflection of reinforced concrete beams, based on observations made by computer vision. Their work supports an innovative method that engineers can use to measure the deflection of reinforced concrete beams. The results of their analysis revealed that the proposed ResNet model that was optimized by JSO was more accurate than conventional ResNet.

Dhevanandhini and Yamuna103 used JSO to find the optimal coefficients of a discrete wavelet transform (DWT) to improve efficient multiple-video watermarking. They analyzed the performance of the proposed method using various metrics and compared it with the DWT-based watermarking approach, which it outperformed.

Elkabbash et al.104 proposed a novel detection system that was based on optimizing the random vector functional link (RVFL) using JSO, following the dimensional reduction of Android application features. They used JSO to determine the optimal configurations of RVFL to improve classification performance. The optimized RVFL minimized the runtime of the models with the best performance metrics.

Gouda et al.105 employed JSO to solve the problem of evaluating the parameters of the polymer exchange membrane fuel cells (PEMFCs) model. The maximum percentage voltage-biased error was ± 1% in all test cases, indicating that JSO can solve this problem more effectively than other algorithms.

Youssef et al.106 used JSO to estimate the parameters of a single-phase power transformer from the current and voltage under any load. They consider difference between the estimated and actual values as the main objective function that must be minimized. Experimental results revealed that the parameters of the transformer equivalent circuit were accurately obtained, indicating that the algorithm can be used to estimate the parameters of a single-phase transformer.

Kızıloluk and Sert107 adopted JSO to optimize the hyper-parameters of the Alex Net CNN model for feature extraction in the Faster R-CNN-JSO model for the early detection of hurricanes from satellite images. The purpose was to alert people about upcoming disasters and thus minimize casualties and material losses. Their results demonstrated that hyper-parameter optimization increased the detection performance of the proposed approach by 10% over that of Alex Net without optimized hyper-parameters. The average precision of Hurricane-faster R-CNN-JS was 97.39%, which was remarkably higher than those of other approaches.

Bisht and Sikander108 used JSO to optimize the parameters of the solar photovoltaic (PV) model. They used JSO to optimize the parameters of a single-diode PV model using various performance measures, such as PV characteristics, power-voltage, and current-voltage curves, relative error (RE), root mean square error (RMSE), mean absolute error (MAE), and normalized mean absolute error (NMAE). Their proposed technique provided better results than other techniques, with a lower RE, RMSE, MAE, and NMAE; it also converged rapidly.

Azam et al.109 utilized JSO to dampen out low-frequency oscillations (LFOs) by tuning the critical parameters of conventional lead-lag type power system stabilizers. JSO is used to tune time-domain simulations of the angular frequency, rotor angle, and control signal. They tested this method on two separate multimachine networks that were exposed to a three-phase fault, and compared it with two well-known optimization algorithms, called PSO and the backtracking search algorithm (BSA). Their results show that JSO provided better damping power system ratio than did the other algorithms. Moreover, the JSO-based approach converged in fewer iterations.

Raja and Periasamy110 presented the block chain and JSO-based deep generative adversarial neural network (DGANN) method for the distributed routing scheme of a wireless sensor network (WSN). They used the block chain routing protocol to detect and store packets and to transfer them from the source to the destination efficiently to improve the security and efficiency of the DGANN method. They used JSO to optimize the weight parameters of the DGANN method. The simulation results demonstrate that in the routing of a WSN, DGANN with optimized parameters outperforms others methods, such as the multidimensional scaling-map, the trust-aware routing protocol through multiple attributes, and dynamic rate-aware classified key distributional secure routing algorithms.

Usharani et al.111 used JSO to optimize the hyperparameters of long short-term memory (LSTM) networks to enhance the error metrics of the approximate multiplier. They used their proposed pre-trained LSTM model to generate approximate design libraries for the different truncation levels as a function of area, delay, power and error metrics. Their experimental results on an 8-bit multiplier with an image processing application reveals that the proposed approximate computing multiplier achieved a superior area and power reduction with very good error rates.

Nyong-Bassey and Epemu112 used JSO and PSO to identify servomechanism parameters using a two-step approach, involving a first-order transfer function and iterative minimization of a fitness score that is derived from the root mean squared error between the experimental and simulated position responses of the servomechanism of an equivalent state-space model structure. The simulated angular position step responses of the servomechanism that runs the JSO and PSO algorithms showed very closely with each other, in terms of root mean squared error. Table 2 summarizes recent advances in the application of the jellyfish search optimizer.

Table 2 Applications of jellyfish search optimizer in various fields.
Next Post

Reporting Issuers Need to be Factual and Balanced, Striving for Accurate and Comprehensive ESG reporting

Reporting Issuers Need to be Factual and Balanced, Striving for Accurate and Comprehensive ESG reporting November 30, 2022 Publications 5 minute read McMillan is authoring a number of bulletins to assist clients in navigating the quickly developing regulatory regime surrounding ESG. The introductory bulletin for the series can be found […]