Abstract
This study delves into the realm of system identification, a crucial sub-field in control engineering, aimed at constructing mathematical models of systems based on input/output data. This work particularly proposes the application of artificial ecosystem algorithm (AEO) for solving system identification problems. Inspired by the energy flow of natural ecosystems, AEO has undergone specific modifications leading to derived versions. Additionally, five diverse meta-heuristic algorithms are employed to assess their applicability and performance in system identification using data from an air stream heater experiment kit. A comprehensive performance comparison is made, considering time bounds, maximum generations, early stopping, and function evaluation constraints, presenting their respective performances. Among the evaluated algorithms, the AEO algorithm enhanced with the sine and cosine strategy stands out with a determined R2 value of 0.951. This algorithm consistently outperforms others in Wilcoxon tests, showcasing its significant success. Our study affirms that meta-heuristic algorithms, particularly the proposed AEO algorithm, can be effectively applied to system identification problems, yielding successful calculations of transfer function parameters.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
The ability to exert control over systems stands as a paramount accomplishment that has propelled humanity to its current technological pinnacle. A prerequisite for effective control is the availability of accurate mathematical models that faithfully capture the essence of the systems. However, it is crucial to acknowledge that modelling these systems can present a formidable challenge, demanding a profound understanding of their intricacies and interplay. System identification methods, initially employed in the defence sector, particularly during the early twentieth century and notably during World War II, have been utilized for obtaining mathematical models from data collected through an experimental setup. Researchers have consistently highlighted the importance of minimizing complexity and maximizing efficiency in determining the transfer function through system identification methods [1].
Several methods have been suggested to obtain the transfer functions using information from both the time and frequency domains. System identification employs various methods, including white box, grey box, and black box, depending on the mathematical model's structure [2, 3]. In the white-box method, a complete understanding of the model is essential. The grey-box method involves indirectly acquiring parameters to construct the model. In the black-box method, the system's information is limited to the input/output data, without any further details available. Although the models derived from the white-box method provide very good results, it is often challenging to have all the information about the system. Therefore, grey-box and black-box methods are much more widely used [4]. Linear grey-box, nonlinear grey-box, and linear polynomial (auto-regressive) models (AR, ARX, ARMA, ARMAX) are employed for situations where the structure of systems is either unknown or partially known [5].
In the context of real-time applications, it is common to employ equipment with constrained resources, such as programmable logic controllers or microcontroller circuits. MATLAB, along with its system identification toolbox, is not suitable for devices with limited resources due to its substantial memory requirements and high cost. Recent advancements in the Python programming language have made it feasible to create real-time algorithms with comparable power to those in MATLAB. Especially, specific distributions like Micro Python are suitable for smaller, more flexible, and embedded systems compared to MATLAB, due to their low resource consumption and microcontroller compatibility. In addition, the Python programming language offers advantages such as a widespread developer community and being available for free licence. Therefore, the Python language has been chosen in the creation of this article. Meta-heuristic algorithms find application in diverse problem-solving scenarios, spanning tasks like optimizing company production processes to designing PI controllers. Consequently, these algorithms exhibit a wide range of practical applications. There have been great advances since these algorithms were first proposed, but many new algorithms are still being proposed. Meta-heuristic algorithms have been employed in addressing numerous single/multi-objective, continuous/discrete, and constrained/unconstrained problems.
One prospective domain for the application of meta-heuristic algorithms involves the parameter estimation of control systems, encompassing both continuous and discrete structures. The Ziegler-Nichols (ZN) open-loop or closed-loop transfer function determination methods partially address the problem in a straightforward manner. However, estimation of the transfer function of many systems is a challenging and time-consuming process. On the other hand, quickly determining the transfer function of a system after collecting the required data from an experimental setup simplifies the controller design process. In this study, research has been conducted to broaden the application scope of meta-heuristic algorithms developed to solve a wide range of problems. In one of the recent studies, meta-heuristic algorithms were used for system identification purposes, and it was stated that they are suitable for obtaining transfer functions depending on input/output data [6]. System identification methods find application in various areas, including boost converter applications. Despite possessing a comprehensive mathematical model, the authors faced challenges in parameterizing the PI (proportional, integral) controller when modelled in Simulink. Nevertheless, they successfully employed system identification methods, yielding satisfactory results in the modelling process. [7, 8].
Zhao et al. [9] proposed a nature-inspired meta-heuristic optimization algorithm called artificial ecosystem optimizer (O-AEO). Hassan et al. [10] suggested a chaotic AEO method for economic emission forecasting. Omotoso et al. [11] proposed an AEO-based optimization algorithm to perform demand-side management in a hybrid energy system. They compared the performance of the AEO algorithm with the performance of Harris hawk optimization (HHO) and the future search algorithm. Accordingly, they reported that the AEO algorithm was more efficient. Mahdad [12] applied the adaptive split AEO algorithm to solve non-smooth economic dispatching by considering some limits. Wireless sensor networks exemplify the practical use of optimization algorithms, particularly in the realm of efficient data transmission, given the constraints imposed by limited resources. A hybrid whale optimization based AEO method has been suggested to improve energy efficiency [13, 14]. Nguyen [15], who stated that reducing power losses in the distribution network is very important, proposed an AEO-based algorithm to reorganize a distribution network. He compared the proposed AEO algorithm with the cuckoo search algorithm and demonstrated that the AEO algorithm is highly effective. Izci et al. [16] developed a hybrid approach by combining the Nelder–Mead (NM) simplex method with the AEO algorithm, which was applied to estimate the proportional, integral, and derivative (PID) parameters of the controller in a DC-DC buck converter. They compared the hybrid method with simulated annealing, whale optimization, and genetic algorithm methods and reported that the performance was quite satisfactory. In another study on PID controller design, the AEO algorithm was used to determine the optimal design parameters of the controller of an automatic voltage regulator [17].
Kumar and Janamala [18] carried out a study based on AEO for the location and size selection of distributed solar photovoltaic systems. They pointed out that the AEO algorithm has very good performance for solving complex, nonlinear, multivariate optimization problems. Güvenç et al. [19] used the AEO method for the coordination of directional overcurrent relays. The authors compared their results against those achieved using differential evolution and opposition-based chaotic differential evolution algorithms, conclusively establishing the proposed algorithm as a potent and effective method for addressing the coordination problem. Similarly, Abdelhamid et al. [20] preferred the AEO method for the coordination of directional overcurrent relays. Shaheen et al. [21] proposed an AEO algorithm for solving the optimal capacitor placement problem in the electricity distribution network. Moussa et al. [22] employed the AEO algorithm to coordinate efficient power scheduling in smart homes and reported satisfactory results in terms of pricing. Niu et al. [23] calculated the adaptive infinite impulse response (IIR) filter parameters using the AEO algorithm in which dynamic opposite learning and nonlinear adaptive weighting coefficient are included. They also applied their algorithm to IIR-based system identification problems.
Menesy et al. [24] adapted their modified AEO (M-AEO) algorithm to the polymer electrolyte membrane fuel cell model. The proposed AEO-based optimization algorithm consists of an operator to ensure the balance between the exploitation and exploration phases. Sultan et al. [25] focus on the optimal design of a grid-connected renewable energy system consisting of photovoltaic, wind turbine, and fuel cells. The proposed system utilizes a fuel cell with a hydrogen gas tank to store energy in chemical form. The optimal component sizes of this hybrid generation system are derived using the improved AEO optimization technique. Eid et al. [26] utilized their AEO-based algorithm to solve optimal placement problems of distributed generation systems. They tried to achieve a balance between the exploitation and exploration phases in their proposed algorithm. In their enhanced AEO (E-AEO) algorithm, they attempted to strike a balance between the exploitation and exploration phases. They integrated the sine–cosine algorithm with the AEO algorithm and presented its performance; the sine–cosine algorithm is a technique that utilizes a mathematical model based on sine and cosine functions [27]. Similarly, Moussa et al. [28] proposed the use of AEO algorithms for reactive power distribution, which is an important field of study for the stable and safe operation of electric power systems. Elkholy et al. [29] carried out a study in which they aimed to develop an efficient tool based on AEO to calculate the unknown model parameters of the system in photovoltaic power generation. They compared the derived parameters with the experimental results and concluded that the performance of AEO is satisfactory. Some other studies on AEO algorithms calculated the parameters of photovoltaic systems and compared with other optimization algorithms [29,30,31,32,33,34]. Barshandeh et al. [35] proposed a hybrid structure where AEO and HHO algorithms are employed together. They applied the proposed algorithm to various engineering problems and common functions and presented its performance. The AEO algorithm has also been utilized in image processing applications as a hybrid with the deep neural network. Sahlol et al. [36] have presented a new hybrid AEO method for the effective classification of X-ray images using chest radiography and deep learning-based image segmentation techniques for tuberculosis diagnosis. Ewees et al. [37] proposed an AEO-based algorithm on the multi-level thresholding method, which is one of the effective image segmentation techniques. Their approach basically combines the differential evolution and AEO methods.
AEO-based algorithms have also been used in internet of things applications. Duhayyim et al. [38] proposed an AEO-based deep-learning model for sustainable waste management. Hosseini et al. [39] used the sine–cosine algorithm and advanced multi-objective hybrid AEO-based optimization algorithm for botnet detection in IoT. Essa et al. [40] proposed the integration of the random vector functional link (RVFL) network with the AEO algorithm to predict the power consumption and water efficiency of a seawater greenhouse. In another study, AEO algorithms have been used to solve the path-planning problem of unmanned combat aircraft [41]. Rizk-Allah and El-Fergany [42], added a nonlinear weight coefficient to the AEO algorithm. They employed this improved AEO (I-AEO) algorithm to define the model parameters of proton exchange membrane fuel cells. One of the derivatives of AEO algorithms was also applied to solve the transient stability constrained optimal power flow problem [43]. Thieu et al. [44] proposed a hybrid artificial intelligence model with a meta-heuristic algorithm to build a monthly groundwater level prediction model. The augmented AEO-based (A-AEO) multilayer perceptron model is constructed on a traditional MLP network, utilizing levy-flight and Gaussian distribution to enhance its optimization capability.
There are more than 80 meta-heuristic algorithms. However, the pursuit of meta-heuristic algorithms with superior performance is rapidly increasing the number of new and modified version of meta-heuristic algorithms. Meta-heuristic algorithms can be categorized as evolutionary, swarm, physics, human, bio, system, and math based (Fig. 1).
Meta-heuristic algorithms, a branch of computer science [48, 49], are designed to tackle large and complex optimization problems, relying on trial-and-error methods. However, the objective of these methods is to approach the solution space with greater effectiveness. This is achieved through the utilization of nature-inspired heuristic or intuitive methods. Likewise, certain machine learning algorithms may encompass various strategies operated with trial-and-error methods [50, 51]. For instance, the reinforcement learning method exhibits a systematic trial-and-error process when searching for solutions [52, 53]. Meta-heuristic algorithms are particularly useful in situations where the equation is known but the optimal values of the parameters are unknown. An illustrative example can be found in the context of designing PID controllers, where numerous studies are devoted to the endeavour of determining the optimal values for the three distinct parameters, namely the proportional gain, integral time, and derivative time, in a commonly known mathematical equation [54,55,56,57].
In order to compare performances of meta-heuristic algorithms in the article, various algorithms, including the AEO algorithm and its variants (O-AEO, I-AEO, E-AEO, M-AEO, A-AEO), as well as the water cycle algorithm (WCA), germinal centre optimization (GCO), grey wolf optimizer (GWO), Harris hawk optimization algorithm (HHOA), and ant lion optimizer (ALO) algorithms have also been considered. In Fig. 1, the selected system-based (green) and swarm-based (orange) algorithms are depicted in different colours. The WCA works by mimicking the natural process of the water cycle, incorporating precipitation, evaporation, and other hydrological phenomena to optimize solution search processes [58]. GCO imitates the competitive and adaptive features observed in the germinal centre reaction within the vertebrate immune system to create an optimization algorithm [59]. The GWO simulates the leadership hierarchy and hunting behaviour of grey wolves, employing hunting, stalking, and encircling strategies to optimize solution search processes [60]. The HHO algorithm is inspired by the hunting behaviour of Harris hawks, utilizing communication, cooperation, and vigilant strategies to optimize solution search processes [61]. The ALO algorithm emulates the trapping behaviour of antlions, employing a combination of random walk and spiral movement to optimize solution search processes [62]. Examining the conducted studies reveals numerous research efforts dedicated to the design of controller parameters using metaheuristic algorithms in the field of control systems. However, limited research has been conducted on deriving transfer functions based on input–output data using meta-heuristic methods. This study contributes to the field of control systems by employing metaheuristic algorithms for system identification through the utilization of the artificial ecosystem algorithm. Our contribution can be summarized as;
-
The AEO algorithms have been employed for the first time in system identification, a subfield of control systems, with the objective of predicting parameters of continuous transfer functions.
-
The efficiency of the AEO algorithms for the system identification problem has been validated through various analyses such as descriptive statistic, nonparametric test, transient response, frequency response, error performance/plot analysis, histogram, and pole zero map. Among the proposed algorithms, the E-AEO algorithm is found to offer better performance, and AEO, in general, can operate with fewer parameters compared to some other meta-heuristic algorithms. Early stopping (ES), function evaluation (FE), time bound (TB), and maximum generations (MG) benchmarks have been presented. The comparison with the differential evolution and opposition-based chaotic differential evolution algorithms has provided clear evidence of the effectiveness and power of the proposed algorithm in addressing the coordination problem.
2 Preliminary Analysis
A PC equipped with an 8-core Intel Core i7-9700 CPU running at 3 GHz, 8 GB of RAM, and a 256 GB SSD was used in acquiring the system model and developing the proposed algorithms. The codes used to predict transfer function parameters were written using Python (3.8) and libraries such as Mealpy 2.1.2 [45], and Control 0.9.2. Jupyter Notebook 3.0.14 were used as integrated development environments (IDEs). The experimental set's data was acquired by utilizing a dataset within MATLAB. These data were derived from the laboratory-scale air stream heater experimental set (Feedback's Process Trainer PT326) shown in Fig. 2. In this experiment set, air passing through a tube is heated at the nozzle, and the air temperature is measured by a thermocouple connected to the tube's outlet. The voltage applied to the inlet heater is denoted as the input reference signal (Volts), while the air temperature, measured by the thermocouple (Volts), is considered the output. The output data contains 1000 discrete measurements from the thermocouple transducer, which converts the temperature value to voltage. The input voltage is sampled randomly from one level to another. The measurement setup has a sampling time of 0.08 s. The model was trained using 80% of the collected data, while the remaining 20% was reserved for the testing phase. Figure 3 illustrates the thermocouple voltage measured at the output in response to the reference voltage applied to the heater input.
3 Methodology
3.1 Artificial Ecosystem Algorithm (O-AEO)
The AEO algorithm was initially proposed by Zhao et al. [9], drawing inspiration from the energy flow within a natural ecosystem. As a concept, an ecosystem describes the environmental relationships between a group of living organisms. This concept is divided into two categories: abiotic and biotic ecosystems. Abiotic ecosystems involve non-living objects such as light, water, and air. However, biotic ecosystems include all living things. AEO is a population-based algorithm that mimics the production, consumption, and decomposition behaviour of organisms. Energy flow and nutrient cycling serve as a crucial balance in maintaining equilibrium in an ecosystem. Living things in an ecosystem are divided into three groups based on their behaviour: producers, consumers, and decomposers. Producers, categorized as green plants, derive their energy from the process of photosynthesis. Consumers are organisms that derive their energy by relying on producers or other consumers (animals). Depending on the type of food, consumers can be categorized into three groups: herbivores which consume only plants, carnivores which feed exclusively on animals, and omnivores which consume both plants and animals. Decomposers, such as bacteria, and fungi play a significant role in converting the remains of an organism's body into molecules. After the decomposition process, the remains are absorbed back into the soil by the producers (plants), and the cycle repeats.
Figure 4 illustrates the energy flow in an example ecosystem. Producers, fed by the sun, are situated at the beginning of the food chain. Consumers encompass a variety of living organisms, making them the most intricate part of the ecosystem. Production is responsible for improving the balance between exploration and exploitation. Consumption is tasked to improve the exploration process, while decomposition is tasked to improve the exploitation process. It is proposed that during the operation of the AEO algorithm, only one producer and one decomposer should be present in each population, while the other individuals should be considered consumers of the three predefined types. However, this proposition may vary depending on the specific requirements and design of the AEO algorithm. The energy level of each individual in the population is determined by the fitness function. The energy flow of the AEO algorithm is depicted in Fig. 5, where producers are highlighted in yellow, whereas decomposers are represented in red. In the algorithm, the worst individual X1 has the highest energy level (producer), while X9 is the best individual (decomposer) with the lowest energy level (fitness value). The other individuals are consumers; X2 and X5 are taken as herbivores, X3 and X7 as omnivores, and X4 and X6 as carnivores.
3.1.1 Production Process
In AEO, the producer mimics the role of the producer in the ecosystem, which needs the decomposer, sunlight, water, and the food supplied by the decomposer to produce food. In AEO, the producer with the lowest fitness value is updated according to the search limits and the best individual (the decomposer). As a consequence of this process, other individuals in the population will update their positions. Using the production operator, a new individual (producer) is generated between the best one (\(x_n\)) and a randomly generated individual (\(x_{{\text{rand}}}\)), expressed mathematically in Eqs. 1–3.
The variables in the equations such as \(n,{\text{max}}_{{\text{iter}}} , U, L,\) and \(r_1\). represent the population size, maximum iteration number, upper limits, lower limits, and a random number ranging between [0, 1]. The coefficients \(a\) and \(r\) represent a linear weight coefficient and a random vector in the range [0, 1].
3.1.2 Consumption Process
After the production process is realized by producers, the consumption process begins with consumers. In this process, each consumer can acquire energy from another consumer or producer linked to a lower energy level. Many animals' foraging mechanisms can be mimicked through a mathematical operator called levy-flight. This operator is also employed to improve the exploration phase in many algorithms. The consumption phase of an algorithm with levy-flight can be denoted by Eqs. 4–5.
where \(N (0,1)\) represents a normal distribution with a mean of 0 and a standard deviation of 1. This consumption factor will help each consumer to obtain food using different hunting strategies. If the consumer is chosen as herbivores at random, then only producers will be consumed. In this case, the consumption behaviour of the consumer (herbivore) will be presented mathematically as follows:
If a consumer is randomly selected as a carnivore, it will eat another consumer with a higher-energy level. In this case, carnivore behaviour can be modelled by Eq. 7.
If an omnivore is selected within the ecosystem, it can eat a consumer or producer with a high energy level. This behaviour is different from the behaviour of herbivores and carnivores. The mathematical model in this case can be represented in Eq. 8.
In Eq. 8, r2 is a random number in the interval [0,1]. The updating of the individual positions is based on random selection or the consumption operator of the most unsuccessful individuals, respectively. In this way, it is possible to search for the global point.
3.1.3 Decomposition Process
This process is essential for AEO algorithms. Decomposers chemically break down deceased producers or consumers, releasing nutrients that are vital for the growth of other producers. In this case, the equation can be stated as Eqs. 9-12
In Eq. 9, the decomposition factor \(D\) and the weight coefficients \(e\) and \(h\) are designed for mathematical modelling of decomposition process. The position represented by \(x_i\) of the ith individual can be determined based on the parameters \(D, e,\) and \(h\), using the position of \(x_n\).
AEO initiates the optimization process by randomly generating a population. At each iteration, the initial search individual updates its position in accordance with Eq. (1), whereas the other individuals have an equal probability of updating their positions using Eqs. (6), (7), or (8). If the calculated function value for the individual is better than the previous one, it is accepted. After this process, the position of each individual is updated according to Eq. 9. The update process in the AEO algorithm continues until the specified maximum number of epochs is reached or until termination criteria are met [9]. The pseudocode for the AEO algorithm is presented in Fig. 6.
3.2 Modified AEO Algorithm
For population-based meta-heuristic algorithms, exploration and exploitation are two important and conflicting phases. A well-defined balance between these two phases is crucial for narrowing the search space and solving the global optimum. The modified version of AEO algorithm by Menesy et al. [24] was used to the Polymer electrolyte membrane fuel cell model. In this study, the \(H\) operator, which is proposed to improve the performance, is represented in Eq. 13 and decreases linearly from 2 to 0 during the iteration time.
In Eq. (13),\({{\text{max}}}_{{\text{iter}}}\) indicates the maximum iteration number. After the addition of operator H to Eqs. 6–8, the new equations are represented as shown in Eqs. 14–16.
3.3 Enhanced AEO Algorithm
Similar to the modified AEO, the enhanced AEO (E-AEO) algorithm proposed by Eid et al. [26], the E-AEO algorithm uses the G operator to balance between exploration and exploitation.
In Eq. 17, \({{\text{max}}}_{{\text{iter}}}\) indicates max. iteration number and \(t\) is the current iteration. In the production phase, this parameter was added to Eq. (6) and new Eq. 18 was formed as follows.
A sine–cosine based algorithm is also proposed in E-AEO algorithm. Equations 6–8 were rearranged as Eqs. 19–21 after the addition of these terms.
In Eqs. 19–21, \({r}_{3}, {r}_{4}, {r}_{5}\) are expressed as \({r}_{3}=2 \pi \times {\text{rand}}\), \({r}_{4}\)(random value between 0 and 1) and \({r}_{5}={\text{rand}}\).
3.4 Improved AEO Algorithm
In most of the meta-heuristic algorithms, there is a weighting coefficient to obtain the appropriate result in the search space during the exploration and exploitation phase. In most algorithms, this weighting algorithm is in linear form. In Rizk-Allah and El-Fergany [42] study, added a nonlinear adaptive weighting coefficient. Accordingly, the adaptive weight (\(a\)) coefficient is as follows:
Equation 22 is actually a derived version of Eq. 2. Additionally, a dynamic crossover strategy has been added to the improved AEO (I-AEO) algorithm. According to this strategy, the improved equations are presented as follows:
Rizk-Allah and El-Fergany [42], who developed the improved version of AEO algorithm, also added the Gaussian strategy to the AEO algorithm to enhance its performance. Gaussian mutation strategy is proposed in some of the meta-heuristic algorithms to improve exploitation capability. In this context, the density function of the Gaussian strategy is given Eq. 25.
where \({\sigma }^{2}\) is defined variance between each individual.
3.5 Augmented AEO Algorithm (A-AEO)
Zhao et al. [9] initially suggest that the AEO algorithm has a strong global search capability; however, its convergence rate diminishes as it approaches the optimal solution due to a higher rate employed in the decomposition phase compared to the rate used in the exploitation phase. Therefore, Nguyen et al. [44] proposed the augmented AEO (A-AEO) algorithm to enhance the exploration (global search) and exploitation (local search) capabilities of the AEO algorithm by using levy-flight and Gaussian (normal) randomization methods together. In the updated consumption phase, the selected agent has a 50% chance of updating its location based on herbivore, omnivore, or carnivore organisms, and a 50% chance of using the proposed levy-flight method. The idea is to use more than one method for a better search in the exploration phase. The consumption phase is presented in Eq. 26.
\(ste{p}^{g+1}\) is generated by the levy distribution with a randomly generated β value in the interval (1,2). In the proposed method, the update process has been revised to enhance the stability of the exploitation phase. In this stage, E factor was replaced by a D-dimensional Gaussian distribution vector. The update used in the algorithm is proposed in Eq. 27. In case the randomly generated value is greater than 0.5, Eq. 28 is taken into account.
where ⊗ is defined as element-wise multiplication.
3.6 System Identification
Black-box modelling is characterized by the iterative process of estimating the coefficients of a proposed mathematical model, with the ultimate aim of identifying the optimal solution through the evaluation of performance results. Initially, a simple linear model is proposed, and if the performance is not sufficient, improvement is achieved by suggesting more complex models [5,6,7]. Equation 29 is a general expression of the black-box model.
In Eq. 29, y(k), u(k), and e(k) are defined as model output, input, and white noise, respectively. G(q) is a transfer function representing the system and H(q) is the disturbance effect. First, the input/output data of the system are collected, some filtering operations are applied, and an appropriate model is selected.
3.6.1 Transfer function model
Transfer functions are defined as a polynomial model that gives the relationship between input reference and output signal. In transfer functions, the degree of the model is equal to the degree of the denominator. The roots of the denominator are called model poles, and the roots of the numerator are called zeros [8]. A transfer function in continuous time is denoted in Eq. 30.
Here Y(s), U(s), and e(s) are defined as output, input, and disturbance, respectively. The \({\text{num}}(s)\) and \({\text{den}}(s)\) given in Eq. 30 are expressed as numerators and denominators describing the relationship between input and output.
3.7 Error Functions
Methods such as integral of square error (ISE), integral of the absolute value of error (IAE), integral of time-weighted squared error (ITSE), and integral of time multiply absolute error (ITAE) are available to determine system parameters and examine their performance [46]. ISE expresses the sum of the squares of the differences between the desired signal generated by a particular control system and the actual system output over time. A lower ISE value indicates better performance of the control system. IAE represents the sum of the absolute values of the differences between the reference signal and the actual system output. Unlike the squared error-based metrics, IAE directly considers the magnitude of the error without squaring it. ITSE involves summing the square of the error at each time step, where the error is typically the difference between the desired reference signal and the actual system output. ITAE is an integral-based metric that evaluates control system performance by considering the product of the absolute error and time at each step. Unlike metrics that solely focus on the absolute error (such as IAE) or squared error (such as ISE), ITAE combines the magnitude of the error with the time duration of the error [8]. The Integral of Time multiplied by Absolute Error (ITAE) serves as a metric in control systems, measuring the accumulated error over time. For this reason, the ITAE metric is proposed to measure the error. Table 1 presents the names and formulas of commonly known error objective functions.
4 Results and Discussions
The performance of the selected AEO algorithms was comprehensively evaluated through comparative analyses with five different algorithms: water cycle algorithm (WCA), germinal centre optimization (GCO), grey wolf optimizer (GWO), Harris hawk optimization algorithm (HHOA), and ant lion optimizer (ALO).The selected algorithms have been used to predict the four parameters (z1, z2, p1, and p2) of the transfer function, with the lower and upper boundary set as [-100, 0, 0, 0] and [100, 100, 100, 20], respectively. Predefined parameters and the classification regarding the algorithms are presented in Table 2.
4.1 Reliability Analysis and Constraints
Although many studies focusing on performance have not paid much attention to stopping criteria, the appropriate choice of stopping criteria is critical for system model parameter estimation. Especially for embedded hardware with constrained resources, it is very important to reach a solution in minimum time [47]. In this study, besides analysing the performance of meta-heuristic algorithms, the results of different stopping criteria are also compared.
4.1.1 Time Bound Constraint Performance
The meta-heuristic algorithms were tested by setting a maximum time bound of 20 s (TB = 20). In order to test the reliability of the algorithms, the independent run number for each algorithm was set to 100. The R2 performance of meta-heuristic algorithm under TB constraint is presented in Fig. 7. This concept implies finding the optimal parameters 100 cycle under the specified criteria. In this case, the success of the algorithms in making accurate predictions can be observed in Fig. 7. However, in cases where the performances are closely matched, it is necessary to conduct descriptive statistical analysis. Descriptive statistics results under TB constraint are presented in Table 3. The mean, minimum, maximum, and standard deviation values are presented. Furthermore, the number of runs with a performance (R2) less than 0.9 after 100 runs is also indicated. The designated value of 0.9 provides insights into the dependability of the algorithms, and its significance lies in the case of it being 0. The E-AEO algorithm achieved the highest R2 value of 0.9698, with an average R2 value of 0.9502. Additionally, the O-AEO (0.9509) and M-AEO (0.9510) algorithms also exhibited satisfied performance.
4.1.2 Maximum Generation Constraint Performance
The R2 values calculated considering the maximum number of generations (MG = 20) are plotted in Fig. 8. Table 4 presents descriptive statistical results under the MG constraint. The GWO algorithm, which has exhibited the highest performance and has never fallen below the 0.9 threshold, has shown successful performance (0.95). However, the I-AEO algorithm has reached a highest R2 value (0.9582) but has a value that falls below 0.9. M-AEO and E-AEO have high average performance values, but have performances fall below 0.9.
4.1.3 Function Evaluation Constraint Performance
The performance results were analysed based on 4000 iterations of the algorithms, taking into account the constraint of function evaluations (FE). Figure 9 demonstrates the performance variations of various algorithms under FE constraints.
Table 5 presents descriptive statistical results under the FE constraint. Considering the criteria stated in the previous section, the O-AEO algorithm has achieved the highest performance (0.9504). Moreover, the results obtained from the M-AEO and GWO algorithm independent run demonstrate notable proficiency. The ALO algorithm has the lowest mean R2 values in this comparison.
4.1.4 Early Stopping Constraint
The early stopping (ES = 3) criterion can be defined as the end of the search process after the minimum calculated error value of the meta-heuristic algorithms remains unchanged for 3 cycles. This criterion gives the algorithms a long time to search for a solution as demonstrated in Fig. 10. Therefore, the meta-heuristic that try to calculate the optimal parameter without getting stuck. Figure 10. shows that the E-AEO algorithm was able to determine the global solution point every time after 100 iterations. Table 6 presents the average, minimum, maximum, count number (< 0.9), and standard deviation value of the algorithms considered in the paper when the ES criterion is taken into account. E-
AEO algorithm achieves the highest performance value (0.954) in the ES constraint when considered together with Table 6 and the Wilcoxon test presented in Table 7. When Table 6 and Table 7 are evaluated together, the E-AEO algorithm attains the highest performance value (0.954) in terms of the ES criterion. The O-AEO (0.951), A-AEO (0.9426), and WCA (0.9460) algorithms also demonstrate high performance values. However, the E-AEO algorithm has surpassed other methods in the Wilcoxon test.
4.2 Nonparametric test analysis
Analyses conducted using the nonparametric statistical method, the Wilcoxon test, are presented in Table 7. The Wilcoxon signed-rank test is a nonparametric test used to determine whether there is a significant difference between the means of two paired samples [63]. In this study, the Wilcoxon test was employed to compare algorithms based on their p-value values and determine the winner. Comparison between the E-AEO algorithm and other algorithms has been considered at a significance level of 5%. The calculated p-values for the considered TB, MG, FE, and ES constraints are provided in Table 7. Additionally, at the end of the Wilcoxon test, the winning algorithms are indicated by a positive sign ( +). It is evident that the E-AEO algorithm has demonstrated superior performance compared to other algorithms under the ES criterion. Moreover, under the TB and MG constraints, it outperformed the other 7 algorithms, and under the FE constraint, it outperformed the other 8 algorithms.
4.3 Variation Analysis of Model Parameters
While developing the model of the system, a transfer function is proposed with 2 coefficients in the numerator and denominator. For simplicity of representation, the numerator of the transfer function was chosen as a first-order term, while the denominator was selected as a second-order term for the system. The proposed second-order transfer function is presented in Eq. 31.
Considering the ES criterion of the E-AEO algorithm, the z1, z2, p1, and p2 values calculated over 100 iterations are given in Fig. 11. If Fig. 11 is analysed, the parameters z1, z2, p1, p2 vary between 12.58–12.605, −1.497–1.492, 12.92–12.89, 6.1–6.122, respectively.
Figure 11 shows that the parameter variations remain nearly constrained throughout the iterations. The negative value of z2 indicates that there is a zero in the positive region on the root locus. If a pole root were in the positive region on the root locus, this would obviously result in instability, but having the zero in the positive region does not cause instability and provides a better representation of the model. Figure 12 shows the pole-zero placement on the root locus plot. Green colour indicates zero and red colour indicate pole placement on root locus. Most of the poles coloured in red are overlapped. Similarly, the green circles shown on the right positive side of the root-locus represent zeros. These overlapping parameters show the similarity of the estimates at a glance. Thus, the pz-map can be presented as evidence that the variation of the parameters is constrained. In conclusion, it can be stated that the E-AEO algorithm is both successful and stable in solving system identification problems. This success can be interpreted that a sine–cosine-based structure is more suitable for solving system identification problems.
4.4 Comparison of Performances Indices
In Sect. 4.2, the number of independent runs was set to 100 to showcase that the algorithms yielded non-random outcomes and to evaluate their performance. In this section, the number of independent runs was set to 1. This concept denotes the execution of algorithms just once cycle to determine the optimal parameters. This allows for conducting the analyses presented in the subsections for the obtained optimal parameters. Table 8 shows the parameters calculated after the MG limit is set to 20. Under the MG constraint, the O-AEO (0.95092), E-AEO (0.95092), I-AEO (0.95099), A-AEO (0.95045), and GWO (0.94927) algorithms have exhibited high performances. In this case, the AEO algorithms except M-AEO have performed more successful optimization compared to the other presented algorithms. The parameter of transfer functions calculated with the AEO algorithms, excluding M-AEO, are very close to each other and the performance indicators MAPE, MAE, MSE, R2 and time are also very close to each other. The ALO (50.08 s) algorithm has exerted more effort for the solution compared to other algorithms.
The ES constraint ensures that the algorithm is stopped for 3 generations once there is no further decrease in the global best solution. Table 9 shows the optimized transfer functions MAPE, MAE, MSE, R2 and time values for meta-heuristic algorithms. When the ES criterion value is set to 3, the performance of all AEO algorithms is quite good, but the solution times increase considerably. Especially, E-AEO has reached the highest performance value (0.95105) while having a solution time of 59.07 s. The performance of M-AEO (0.95083) and A-AEO (095047) algorithms is also high; however, their respective solution times of 218.79 s. and 11.725 s. are much longer compared to E-AEO.
In the FE constraint criterion, the running algorithm is halted when it reaches the given maximum function number. The termination criterion used in this case closely resembles the MG stopping criterion. Throughout the study, a maximum constraint value of 4000 was set for fitness evaluations (FE), taking into account the dataset's distinct characteristics. Table 10 presents the performance indicators calculated under these specific conditions, illustrating the optimized transfer functions and performance metrics achieved through the utilization of meta-heuristic algorithms. Under the FE criterion, the E-AEO algorithm has achieved the highest performance (0.95117) with a solution time of 23.17 s. The M-AEO (0.80411), GCO (0.76997), and ALO (0.72665) algorithms have the lowest performance. The other algorithms perform adequately. The FE constraint enables the algorithms to solve within similar time bounds. The high performance of four different AEO algorithms under the FE criterion is important in demonstrating the effectiveness of the AEO.
The TB constraint is essentially based on stopping the algorithm at the end of the given time, regardless of the performance result. In this study, considering the characteristics of the data and the structure of the algorithms, the time limit was set to 20 s. In this case, the transfer functions and performance values are presented in Table 11. In accordance with Table 11, the I-AEO algorithm has demonstrated the best performance (0.95123). The algorithm with the lowest performance (0.67424) has been identified as ALO. Among the AEO algorithms, M-AEO has the lowest performance (0.72221).
The algorithms considered in problem-solving have been examined in terms of performance, solution times, and variations in transfer function parameters. Upon examining the presented statistical results, graphs, Wilcoxon test, and performance criteria, it is evident that AEO algorithms are quite sufficient in solving system identification problems. Among the AEO algorithms, particularly the E-AEO algorithm has stood out compared to other algorithms. Furthermore, for illustrative purposes, the changes in exploration and exploitation are also addressed in Fig. 13. The exploration and exploitation change of the E-AEO algorithm under the TB constraint is presented in Fig. 13a. Also, Fig. 13b shows the global best change, which indicates that the solution is almost found after approximately the 13th iteration.
4.5 Time and Frequency Response Analysis
The time and frequency response of the optimized transfer functions is important for making an interpretation on behalf of the system. At this point, based on the TB constraint, the system response of the relevant transfer functions to the step input is presented in Fig. 14. In Fig. 14, O-AEO, I-AEO, E-AEO, and M-AEO algorithms have almost the same time response. It is difficult to comprehend the transient response performance through Fig. 14; therefore, Table 12 has been provided. In this table, considering the transient response of the optimized transfer functions, it can be observed that although there are slight differences in the rise times, the settling times of the AEO algorithms are rapid. Among the AEO algorithms, the E-AEO algorithm has the lowest settling time (1.30 s.) and peak time (1.71 s.).
Table 12 presents the transient values such as rise time, settling time, overshoot, undershoot, peak value, and peak time related to the transfer functions derived using meta-heuristic algorithms. Figure 15 shows the frequency response (bode graph) of the transfer functions estimated by meta-heuristic algorithms. The bode diagram helps determine the system's gain and phase margins, which are key indicators of stability. If the gain or phase margin is too low, it suggests that the system may become unstable or exhibit undesired oscillations.
If the frequency response is carefully observed, all algorithms, except the GCO algorithm, exhibit a similar frequency amplitude and phase, and the transfer functions obtained using meta-heuristic algorithms demonstrate stable behaviour. In Fig. 16, the error between the predicted transfer function response for each meta-heuristic algorithm and the original output data is presented. This figure indicates that the errors are within a constrained range. The main reason for the variability in error values in Fig. 16 is the variations in the original dataset. The error values being close to 0 indicate the prediction of a strong model.
Figure 17 presents a histogram of the errors. In the histogram figure, the narrow range of error bars indicates that the performances are satisfactory. Drawing conclusions based on error plots is a relatively challenging task. Therefore, the histogram of errors is essential for observing the errors of the obtained transfer functions. The histogram of O-AEO, E-AEO, M-AEO, I-AEO, and GWO algorithms is found within a narrow range. Algorithms observed within a narrow range in the histogram have less error compared to other algorithms.
Figure 18 presents the voltage amplitude versus time of the air stream dryer and the predicted transfer functions. The transfer functions produced by the WCA, GWO, GCO and M-AEO algorithms track the original data with a larger error while the O-AEO, I-AEO, E-AEO, and A-AEO successfully track the data.
The presented graphs depict the comparison of meta-heuristic algorithms and demonstrate their performance under different types of constraints. To test the problem-solving abilities of the AEO algorithms, the reliability of the meta-heuristic algorithms was analysed by checking their R2 values after independent run number for each algorithm was set to 100. Additionally, after setting the independent run number to 1, optimized transfer functions were presented, and performance indicators such as transient response and frequency response were examined. Several analyses have demonstrated that the E-AEO algorithm has achieved higher success compared to other AEO and meta-heuristic algorithms. In the Wilcoxon test, the E-AEO algorithm has outperformed other algorithms under the ES criterion. This success demonstrates that meta-heuristic algorithms especially AEO algorithm can be easily applied to system identification problems.
5 Conclusion
The primary focus of this article is a research investigation that utilizes artificial ecosystem, water cycle, germinal centre, grey wolf, and Harris hawk optimizer algorithms for black-box system identification. The study involves processing reference signal and outlet temperature data obtained from the air stream heater. Algorithms have been compared under four different constraints (time bound, maximum generation, function evaluation, and early stopping constraints). Furthermore, reliability tests were conducted by examining descriptive statistics, the Wilcoxon nonparametric test, transient/frequency responses, and error performances.
It was shown that artificial ecosystem, and 5 state-of-the art optimization algorithms can be easily applied in system identification problems. Artificial ecosystem algorithms have exhibited more successful performance compared to other meta-heuristic algorithms. Under the early stopping criterion, the enhanced artificial eco system algorithm outperformed other algorithms. The mean R2 performances of the time bound, maximum generation, function evaluation, and early stopping constraints are 0.9502, 0.9425, 0.9466 and 0.954, respectively. When the parameter variation and pole-zero map of the transfer functions optimized with the artificial ecosystem algorithm under the early stopping criterion were examined, it was observed that stable and similar solutions were generated in each iteration. The algorithms were compared through the Wilcoxon test, and the E-AEO algorithm demonstrated superior performance based on the obtained p-values.
Data Availability
Data will be made available on request.
References
Zadeh, L. A.: The role of fuzzy logic in modeling, identification and control. In: Fuzzy Sets, Fuzzy Logic, and Fuzzy Systems: Selected Papers by Lotfi A Zadeh. pp. 783–795. (1996).
Manganiello, P.; Ricco, M.; Monmasson, E.; Petrone, G.; Spagnuolo, G.: On-line optimization of the P&O MPPT method by means of the system identification, In: IECON 2013 - 39th Annual Conference of the IEEE Industrial Electronics Society, pp. 1786–1791, (2013). doi: https://doi.org/10.1109/IECON.2013.6699402.
Baždarić, R.; Matko, D.; Leban, A.; Vončina, D.; Škrjanc, I.: Fuzzy model predictive control of a DC-DC boost converter based on non-linear model identification. Math. Comput. Model. Dyn. Syst.Comput. Model. Dyn. Syst. 23(2), 116–134 (2017). https://doi.org/10.1080/13873954.2016.1232283
Ruisheng, L.; Armstrong, M.; Gadoue, S.; Wang, C.: On-line parameter estimation of non-minimum phase switch mode power DC-DC boost converters, In: IET Conference Proceedings, 2016, pp. 6-6. doi: https://doi.org/10.1049/cp.2016.0172 IET Digital Library.
Celikel, R.; Gundogdu, A.: System identification-based MPPT algorithm for PV systems under variable atmosphere conditions using current sensorless approach. Int. Trans. Elect. Energy Syst. (2020). https://doi.org/10.1002/2050-7038.12433
Zaloğlu, M.; Şehmus F.; Erkan, E.: Meta Heuristik Optimizasyon Algoritmalarının Sistem Tanımlama Problemine Uygulanması ve Performans Karşılaştırması. In: International Conference on Engineering, Natural and Social Sciences, vol. 1, pp. 510- 515. (2023).
Fidan, Ş.; Erkan, E.: Boost Konvertörün Black-Box Sistem Tanımlama Yöntemi ile Transfer Fonksiyonunun Elde Edilmesi ve Parçacık Sürü Algoritması Tabanlı PI Kontrolör Tasarımı. In: International Conference on Recent Academic Studies Vol. 1, No. 1, pp. 210–217. (2023)
Fidan, Ş.; Sevim, D.; Erkan, E.: System Identification and Control of High Voltage Boost Converter, In: 2022 Global Energy Conference (GEC), Batman, Turkey, pp. 25-31. (2022)1 doi: https://doi.org/10.1109/GEC55014.2022.9986621
Zhao, W.; Wang, L.; Zhang, Z.: Artificial ecosystem-based optimization: a novel nature-inspired meta-heuristic algorithm. Neural Comput. Appl.Comput. Appl. 32(13), 9383–9425 (2020). https://doi.org/10.1007/s00521-019-04452-x
Hassan, M.H.; Kamel, S.; Salih, S.Q.; Khurshaid, T.; Ebeed, M.: Developing chaotic artificial ecosystem-based optimization algorithm for combined economic emission dispatch. IEEE Access 9, 51146–51165 (2021). https://doi.org/10.1109/ACCESS.2021.3066914
Omotoso, H.O.; Al-Shaalan, A.M.; FarhAl-Shamma’a, H.M.H.A.A.: Techno-economic evaluation of hybrid energy systems using artificial ecosystem-based optimization with demand side management. Electronics 11(2), 204 (2022). https://doi.org/10.3390/electronics11020204
Mahdad, B.: Adaptive split artificial ecosystem-based optimization to solving non-smooth economic dispatch. Transactions of the Indian National Academy of Engineering 7(3), 873–895 (2022). https://doi.org/10.1007/s41403-022-00334-2
Devi, R. R.; Sethukarasi, T.: HIWAEO: Hybrid Improved Whale Artificial Ecosystem Optimization Algorithm based Energy-Efficient Routing Protocol for Wireless Sensor Network. (2021) Doi: https://doi.org/10.21203/rs.3.rs-305087/v1
Devi, R.R.; Sethukarasi, T.: An energy-efficient routing based on a hybrid improved whale artificial ecosystem optimization algorithm in WSN. Concurren. Comput. Practice Exper. 34(4), e6639 (2022)
Nguyen, T.T.: A novel metaheuristic method based on artificial ecosystem-based optimization for optimization of network reconfiguration to reduce power loss. Soft. Comput.Comput. 25(23), 14729–14740 (2021). https://doi.org/10.1007/s00500-021-06346-4
Izci, D.; Hekimoğlu, B.; Ekinci, S.: A new artificial ecosystem-based optimization integrated with Nelder-Mead method for PID controller design of buck converter. Alex. Eng. J. 61(3), 2030–2044 (2022). https://doi.org/10.1016/J.AEJ.2021.07.037
Ćalasan, M.; Micev, M.; Djurovic, Ž.; Mageed, H. M. A.: Artificial ecosystem-based optimization for optimal tuning of robust PID controllers in AVR systems with constrained value of excitation voltage. Int. J. Elect. Eng. Edu., 0020720920940605. (2020)
Kamal Kumar, U.; Janamala, V.: Artificial Ecosystem-Based Optimization for Optimal Location and Sizing of Solar Photovoltaic Distribution Generation in Agriculture Feeders. pp. 743–757. (2022) doi: https://doi.org/10.1007/978-981-16-9416-5_55
Guvenc, U.; Bakir, H.; Duman, S.: Optimal Coordination of Directional Overcurrent Relays Using Artificial Ecosystem-Based Optimization. pp. 150–164. (2021) doi: https://doi.org/10.1007/978-3-030-79357-9_15
Abdelhamid, M.; Kamel, S.; Mohamed, M. A.; Rahmann, C.: An Effective Approach for Optimal Coordination of Directional Overcurrent Relays Based on Artificial Ecosystem Optimizer. In: 2021 IEEE International Conference on Automation/XXIV Congress of the Chilean Association of Automatic Control (ICA-ACCA), pp. 1–6. (2021). doi: https://doi.org/10.1109/ICAACCA51523.2021.9465189
Shaheen, A.; Elsayed, A.; Ginidi, A.; El-Sehiemy, R.; Elattar, E.: Reconfiguration of electrical distribution network-based DG and capacitors allocations using artificial ecosystem optimizer: Practical case study. Alex. Eng. J. 61(8), 6105–6118 (2022). https://doi.org/10.1016/j.aej.2021.11.035
Mouassa, S.; Tostado-Véliz, M.; Jurado, F.: Efficient power scheduling in smart homes using a novel artificial ecosystem optimization technique considering two pricing schemes. Int. J. Emerg. Electr. Power Syst.Emerg. Electr. Power Syst. 22(6), 643–660 (2021). https://doi.org/10.1515/ijeeps-2021-0104
Niu, Y.; Yan, X.; Wang, Y.; Niu, Y.: Dynamic opposite learning enhanced artificial ecosystem optimizer for IIR system identification. J. Supercomput.Supercomput. 78(11), 13040–13085 (2022). https://doi.org/10.1007/s11227-022-04367-w
Menesy, A.S.; Sultan, H.M.; Korashy, A.; Banakhr, F.A.; Ashmawy, M.G.; Kamel, S.: Effective parameter extraction of different polymer electrolyte membrane fuel cell stack models using a modified artificial ecosystem optimization algorithm. IEEE Access 8, 31892–31909 (2020). https://doi.org/10.1109/ACCESS.2020.2973351
Sultan, H.M.; Menesy, A.S.; Kamel, S.; Korashy, A.; Almohaimeed, S.A.; Abdel-Akher, M.: An improved artificial ecosystem optimization algorithm for optimal configuration of a hybrid PV/WT/FC energy system. Alex. Eng. J. 60(1), 1001–1025 (2021). https://doi.org/10.1016/j.aej.2020.10.027
Eid, A.; Kamel, S.; Korashy, A.; Khurshaid, T.: An enhanced artificial ecosystem-based optimization for optimal allocation of multiple distributed generations. IEEE Access 8, 178493–178513 (2020). https://doi.org/10.1109/ACCESS.2020.3027654
Mirjalili, S. M.; Mirjalili, S. Z.; Saremi, S.; Mirjalili, S.: Sine cosine algorithm: theory, literature review, and application in designing bend photonic crystal waveguides. Nature-inspired optimizers, 201–217. (2020)
Mouassa, S.; Jurado, F.; Bouktir, T.; Raja, M.A.Z.: Novel design of artificial ecosystem optimizer for large-scale optimal reactive power dispatch problem with application to Algerian electricity grid. Neural Comput. Appl.Comput. Appl. 33(13), 7467–7490 (2021). https://doi.org/10.1007/s00521-020-05496-0
Elkholy, M.M.; El-Hameed, M.A.; El-Fergany, A.A.: Artificial ecosystem-based optimiser to electrically characterise PV generating systems under various operating conditions reinforced by experimental validations. IET Renew. Power Gener.Gener. 15(3), 701–715 (2021). https://doi.org/10.1049/rpg2.12059
Diab, A.A.Z.; Ezzat, A.; Rafaat, A.E.; Denis, K.A.; Abdelsalam, H.A.; Abdelhamid, A.M.: Optimal identification of model parameters for PVs using equilibrium, coot bird and artificial ecosystem optimisation algorithms. IET Renew. Power Gener.Gener. 16(10), 2172–2190 (2022). https://doi.org/10.1049/rpg2.12510
El-Dabah, M.A.; El-Sehiemy, R.A.; Becherif, M.; Ebrahim, M.A.: Parameter estimation of triple diode photovoltaic model using an artificial ecosystem-based optimizer. Int. Trans. Elect. Energy Syst. (2021). https://doi.org/10.1002/2050-7038.13043
Nguyen, T.T.; Nguyen, T.T.; Tran, T.N.: Parameter estimation of photovoltaic cell and module models relied on metaheuristic algorithms including artificial ecosystem optimization. Neural Comput. Appl.Comput. Appl. 34(15), 12819–12844 (2022). https://doi.org/10.1007/s00521-022-07142-3
Yousri, D.; Babu, T.S.; Mirjalili, S.; Rajasekar, N.; Elaziz, M.A.: A novel objective function with artificial ecosystem-based optimization for relieving the mismatching power loss of large-scale photovoltaic array. Energy Convers. Manage. 225, 113385 (2020). https://doi.org/10.1016/j.enconman.2020.113385
Yousri, D.; Rezk, H.; Fathy, A.: Identifying the parameters of different configurations of photovoltaic models based on recent artificial ecosystem-based optimization approach. Int. J. Energy Res. 44(14), 11302–11322 (2020). https://doi.org/10.1002/er.5747
Barshandeh, S.; Piri, F.; Sangani, S.R.: HMPA: an innovative hybrid multi-population algorithm based on artificial ecosystem-based and Harris Hawks optimization algorithms for engineering problems. Eng. Comput.Comput. 38(2), 1581–1625 (2022). https://doi.org/10.1007/s00366-020-01120-w
Sahlol, A.T.; AbdElaziz, M.; Tariq Jamal, A.; Damaševičius, R.; Farouk Hassan, O.: A novel method for detection of tuberculosis in chest radiographs using artificial ecosystem-based optimisation of deep neural network features. Symmetry 12(7), 1146 (2020). https://doi.org/10.3390/sym12071146
Ewees, A.A.; Abualigah, L.; Yousri, D.; Sahlol, A.T.; Al-qaness, M.A.A.; Alshathri, S.; Elaziz, M.A.: Modified artificial ecosystem-based optimization for multilevel thresholding image segmentation. Mathematics 9(19), 2363 (2021). https://doi.org/10.3390/math9192363
Al Duhayyim, M.; Mohamed, H.G.; Aljebreen, M.; Nour, M.K.; Mohamed, A.; Abdelmageed, A.A.; Yaseen, I.; Mohammed, G.P.: Artificial ecosystem-based optimization with an improved deep learning model for IoT-assisted sustainable waste management. Sustainability 14(18), 11704 (2022). https://doi.org/10.3390/su141811704
Hosseini, F.; Gharehchopogh, F.S.; Masdari, M.: MOAEOSCA: an enhanced multi-objective hybrid artificial ecosystem-based optimization with sine cosine algorithm for feature selection in botnet detection in IoT. Multimed. Tools Appl. (2022). https://doi.org/10.1007/s11042-022-13836-6
Essa, F.A.; AbdElaziz, M.; Elsheikh, A.H.: Prediction of power consumption and water productivity of seawater greenhouse system using random vector functional link network integrated with artificial ecosystem-based optimization. Process. Saf. Environ. Prot.Saf. Environ. Prot. 144, 322–329 (2020). https://doi.org/10.1016/j.psep.2020.07.044
Niu, Y.; Yan, X.; Wang, Y.: An adaptive neighborhood-based search enhanced artificial ecosystem optimizer for UCAV path planning. (2022) doi: https://doi.org/10.1016/j.eswa.2022.118047
Rizk-Allah, R.M.; El-Fergany, A.A.: Artificial ecosystem optimizer for parameters identification of proton exchange membrane fuel cells model. Int. J. Hydrogen Energy 46(75), 37612–37627 (2021). https://doi.org/10.1016/j.ijhydene.2020.06.256
Sonmez, Y.; Duman, S.; Kahraman, H.T.; Kati, M.; Aras, S.; Guvenc, U.: Fitness-distance balance based artificial ecosystem optimisation to solve transient stability constrained optimal power flow problem. J. Exp. Theor. Artif. Intelli. (2022). https://doi.org/10.1080/0952813X.2022.2104388
Van Thieu, N.; Barma, S.D.; Van Lam, T.; Kisi, O.; Mahesha, A.: Groundwater level modeling using augmented artificial ecosystem optimization. J. Hydrol.Hydrol. 617, 129034 (2023)
Van Thieu, N.; Mirjalili, S.: MEALPY: An open-source library for latest meta-heuristic algorithms in Python. J. Syst. Architect. 139, 102871 (2023)
Awouda, A.; Mamat, R.: New PID tuning rule using ITAE criteria. Int. J. Eng. 3(6), 597–608 (2010)
Ravber, M.; Liu, S.H.; Mernik, M.; Črepinšek, M.: Maximum number of generations as a stopping criterion considered harmful. Appl. Soft Comput.Comput. 128, 109478 (2022)
Zhang, W.; Pan, K.; Li, S.; Wang, Y.: Special Forces Algorithm: A novel meta-heuristic method for global optimization. Math. Comput. Simul. (2023).
Yang, X. S.; Chien, S. F.; Ting, T. O.: Computational intelligence and metaheuristic algorithms with applications. Sci. World J. (2014).
Kang, S.; Jin, R.; Deng, X.; Kenett, R. S:. Challenges of modeling and analysis in cybermanufacturing: a review from a machine learning and computation perspective. J. Intell. Manuf., 1–14. (2021).
Dayhoff, J.E.; DeLeo, J.M.: Artificial neural networks: opening the black box. Cancer: Interdiscip. Inte. J. Am. Cancer Soci. 91(S8), 1615–1635 (2001)
Booth, S.; Knox, W.B.; Shah, J.; Niekum, S.; Stone, P.; Allievi, A.: The perils of trial-and-error reward design: misdesign through overfitting and invalid task specifications. Proce. AAAI Conf. Artif. Intell. 37(5), 5920–5929 (2023)
Barto, A. G.: Reinforcement learning. In Neural systems for control. pp. 7–30. Academic Press. (1997)
Joseph, S. B.; Dada, E. G.; Abidemi, A.; Oyewola, D. O.; Khammas, B. M.: Metaheuristic algorithms for PID controller parameters tuning: Review, approaches and open problems. Heliyon. (2022)
Goud, H.; Swarnkar, P.: Investigations on metaheuristic algorithm for designing adaptive PID controller for continuous stirred tank reactor. Mapan 34(1), 113–119 (2019)
Valluru, S.K.; Singh, M.: Optimization strategy of bio-inspired metaheuristic algorithms tuned PID controller for PMBDC actuated robotic manipulator. Proc. Comput. Sci. 171, 2040–2049 (2020)
Ulusoy, S.; Nigdeli, S.M.; Bekdaş, G.: Novel metaheuristic-based tuning of PID controllers for seismic structures and verification of robustness. J. Build. Eng. 33, 101647 (2021)
Eskandar, H.; Sadollah, A.; Bahreininejad, A.; Hamdi, M.: Water cycle algorithm–A novel metaheuristic optimization method for solving constrained engineering optimization problems. Comput. Struct.. Struct. 110, 151–166 (2012)
Villaseñor, C.; Arana-Daniel, N.; Alanis, A.Y.; López-Franco, C.; Hernandez-Vargas, E.A.: Germinal center optimization algorithm. Int. J. Comput. Intell. Syst. 12(1), 13–27 (2018)
Mirjalili, S.; Mirjalili, S.M.; Lewis, A.: Grey wolf optimizer. Adv. Eng. Softw.Softw. 69, 46–61 (2014)
Heidari, A.A.; Mirjalili, S.; Faris, H.; Aljarah, I.; Mafarja, M.; Chen, H.: Harris hawks optimization: Algorithm and applications. Futur. Gener. Comput. Syst.. Gener. Comput. Syst. 97, 849–872 (2019)
Mirjalili, S.: The ant lion optimizer. Adv. Eng. Softw.Softw. 83, 80–98 (2015)
Divine, G.; Norton, H.J.; Hunt, R.; Dienemann, J.: A review of analysis and sample size calculation considerations for Wilcoxon tests. Anesth. Analg. 117(3), 699–710 (2013)
Funding
Open access funding provided by the Scientific and Technological Research Council of Türkiye (TÜBİTAK).
Author information
Authors and Affiliations
Contributions
Şehmus Fidan was involved in the conceptualization, methodology, investigation, writing—original draft and writing—review and editing.
Corresponding author
Ethics declarations
Conflict of Interest
The author declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Fidan, Ş. Artificial Ecosystem Optimizer-Based System Identification and Its Performance Evaluation. Arab J Sci Eng (2024). https://doi.org/10.1007/s13369-024-08841-w
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s13369-024-08841-w