Improved Fitness-Dependent Optimizer for Solving Economic Load Dispatch Problem

Economic load dispatch depicts a fundamental role in the operation of power systems, as it decreases the environmental load, minimizes the operating cost, and preserves energy resources. The optimal solution to economic load dispatch problems and various constraints can be obtained by evolving several evolutionary and swarm-based algorithms. The major drawback to swarm-based algorithms is premature convergence towards an optimal solution. Fitness-dependent optimizer is a novel optimization algorithm stimulated by the decision-making and reproductive process of bee swarming. Fitness-dependent optimizer (FDO) examines the search spaces based on the searching approach of particle swarm optimization. To calculate the pace, the fitness function is utilized to generate weights that direct the search agents in the phases of exploitation and exploration. In this research, the authors have used a fitness-dependent optimizer to solve the economic load dispatch problem by reducing fuel cost, emission allocation, and transmission loss. Moreover, the authors have enhanced a novel variant of the fitness-dependent optimizer, which incorporates novel population initialization techniques and dynamically employed sine maps to select the weight factor for the fitness-dependent optimizer. The enhanced population initialization approach incorporates a quasi-random Sabol sequence to generate the initial solution in the multidimensional search space. A standard 24-unit system is employed for experimental evaluation with different power demands. The empirical results obtained using the enhanced variant of the fitness-dependent optimizer demonstrate superior performance in terms of low transmission loss, low fuel cost, and low emission allocation compared to the conventional fitness-dependent optimizer. The experimental study obtained 7.94E−12, the lowest transmission loss using the enhanced fitness-dependent optimizer. Correspondingly, various standard estimations are used to prove the stability of the fitness-dependent optimizer in phases of exploitation and exploration.


Introduction
After the development of computers, the main objective was to investigate unknown solutions and find the best possible solution. During World War II, Alan Turing broke a cipher of Germany named Enigma by using an algorithm used for searching [1]. Many challenges arose in solving problems in real life due to the improvements in working methods and the exciting acceleration in the extent of computations. Hence, techniques based on numerical programming and conventional logic emerged to overcome the drawbacks of instantly and capably resolving complicated problems [2]. Various algorithms like optimization problems have been designed to manage these limitations. e best possible solution was gained through the optimization method by studying its parameter. All the possible values of present solutions were expressed as a set one of which is the fittest solution. Usually, problems of optimization are solved to design algorithms of optimization [3]. Optimization algorithms are classified into two groups: stochastic algorithms and deterministic algorithms [4,5]. Deterministic algorithms generate a group of related answers when the iterations are started by using an introductory initial point all this happened by using inclination [6]. On the other hand, stochastic algorithms constantly generate distinct answers with related values in the absence of inclination. Diversely, concluding values have a slight difference. ere are two categories of stochastic algorithms: metaheuristic and heuristic [7,8].
Heuristic algorithms use the trial-and-error method to find a solution, and it is supposed that these algorithms will consume reasonable time to reach a solution [9,10]. Moreover, the aim of heuristic algorithms is to utilize various methods in local examinations and techniques of randomization [11]. Further analysis and advancements were made in heuristic algorithms and converted to metaheuristic algorithms [12,13]. e novel collections of algorithms have better performance than heuristic algorithms; accordingly, the affix of "meta" that means "far off" was linked with these algorithms.
Recently, available problems of the real world have turned complicated in considerations of cost, time, and space; it is not possible to traverse all credible solutions. Hence, fast and low-cost techniques are required [14,15]. us, scientists studied the natural events and behaviors of animals to resolve these issues, like how ants select their paths; how fish, flies, or birds chase their prey; and the working of gravity. So, all the algorithms inspired by nature are called nature-inspired algorithms [16]. FDO algorithm that is also known as the fitness-dependent optimizer was introduced by Jaza Abdullah and Tarik Rashid in 2019. e FDO algorithm studies the bee swarms' reproduction practices and follows the activities of swarms. is algorithm finds out the best solution among the pool of solutions [17]. Intelligent computations are rising in various areas of research because of their capability to integrate with large complex and interconnected systems with high speed and accuracy.
Economic load dispatch (ELD) is the most vital and significant field of power system planning and operation [18,19]. e chief aim of ELD was to list a group of real power provided by resources of online generation to satisfy the lacked demand whenever needed under a group of limitations [20], regarding system and unit technical constraints with the least production cost. In general, the ELD problem can be validated as an extremely nonlinear and nonsmooth stifled problem of optimization usually for huge systems. e cost of fuel is concerned with the varying costs of the generation of electricity. e main purpose of addressing the ELD problem is to determine the necessary output power to satisfy the system's requirements in such a way that the cost is limited to its possible minimum value and limits, such as prohibited operating zone (POZ) and valve-point effects (VPEs) [20].
ere is a need for efficient ways of producing electricity. Increasing the cost of fuel makes the method of power production expensive. Advanced systems are therefore expected to propose an economical generation, delivery, and transmission method while keeping electrical limitations in mind. e total device requirement is divided into several units by ELD, which reduces the total cost of generation. Several complications exist while obtaining the ELD problem's global best solution; the results are less accurate due to the nonlinear nature of classical methods, confined to convergence issues, and the best local solutions [6]. e major drawback of some techniques like evolutionary computing is premature convergence [21]. In heuristics, exploitation, and exploration perform a significant role. e capacity of an algorithm to hunt globally is called exploration, and the ability to search locally is known as exploitation [22]. e stability of exploration and exploitation highly affects the swarm-based algorithms' performance. Smaller exploration and extreme exploitation lead to premature convergence; and on the other hand, more limited exploitation and higher exploration can cause barriers to gaining the best solution [23].
In this research, by reducing fuel expense, emission allocation, and transmission loss, the authors have implemented FDO to solve the ELD problem. Similarly, the authors have enhanced a new FDO variant that consolidates specific techniques of population initialization and manipulates sinus maps to pick the FDO weight factor dynamically. e enhanced population initialization method combines a quasi-random Sabol sequence to generate the initial solution for the multidimensional search space. A typical 24-unit device is implemented with different power demands for preliminary evaluation. e observational results obtained by implementing the enhanced FDO variant illustrate the outstanding performance compared to the standard FDO in low transmission loss, low fuel cost, and low emission allocation.

Related Work
In this section, various studies have been reported similar to the enhanced approach but with different evolutionary and swarm-based algorithms, that is, PSO, BA, DE, GA, BCO, and other combinations of the recent state-of-the-art algorithms.
e authors carried out a study to suggest a method for the solution of ELD problems and to contrast it with various available solutions [24]. eir enhanced method owns the following characteristics, managing issues of nondifferentiability, all the problems caused by PBC are also resolved, and problems of multi-objective nature of PBC are also resolved. ey implemented their approach on 5 generation systems, and the achieved outcomes proved that more effective Pareto-curve is gained. An enhanced algorithm inspired by nature, that is, Bat algorithm that offers firm convergence and excellent computational performance, was conducted by [7]. Yang enhanced the BAT algorithm in 2010 after inspiring by bat's echolocation behavior. ey stated that this bat quality enables the bat to locate the prey, that is, various insects, even in the absence of light. eir enhanced method aimed at reducing the total cost of generation in the case of a thermal power plant.
In another research, an enhanced version of GA by utilizing mutation and crossover for the solution of CHPED problems was adopted by [25]. ey stated that primary GA is grown in three phases. In their first phase, they did not include the process of selection to bypass population diversity loss, while in the second phase, they utilized two various crossover operations to dig data about parents and produce possible children. In the third phase, they used the operation of mutation to substitute children with children of other parents. ey proved that their enhanced algorithm is the best substitute for the CHPED problem. Another study [26] was carried out to propose a novel quantum bat algorithm (QBA) based on quantum computing; its main purpose was to solve the problem of multi-objective combined economic emission dispatch (CEED). To minimize the system's nonlinearities, they represented CEED by utilizing the function of the cubic criterion. eir primary concern was the eruption of CO2, NOx, and NOx and load dispatch.
us, it is known as a multi-objective problem of optimization. eir outcomes proved that QBA is the best solution for the problem of CEED as compared to different available solutions. An improved self-adaptable differential evolution algorithm integrating with multiple mutation strategies (ADE-MMSs) as a solution to ELD problems [27]. ey suggested a strategy to improve and explore of basic DE problem. eir enhanced method has 3 expansions of DE. Furthermore, they suggested an approach to manage constraints of equality of problems of ED. eir algorithm enhances the speed of convergence as well as maintains a balance between exploration and exploitation. ADE-MMS is proved by them to be the most suitable solution.
Novel differential evolution algorithm was enhanced in a study [28], for the solution of simultaneous power flow OARPD problems for renewable generators. eir enhanced algorithm, that is, DEa-AR, utilized a combination of arithmetic crossover and performed scaling based on Laplace distribution. For the evaluation of their approach, they utilized the IEEE 57-bus system in various situations. e outcomes of their simulations verified that the suggested method could be used for solving OARPD problems with sources that are efficiently renewable and can provide optimum solutions. Qiao and Liu [29] have carried out to propose a combined framework of EVs and wind farms (WEVs) that reduced the over and underestimation of wind power by utilizing the discharging and charging capability of EVs. ey designed a dynamic economic emission dispatching based on the WEV system (WE_DEED). ey utilized their algorithm for the solution of complicated problems of WE_DEED. While they handle the limitations of WE_DEED through their enhanced algorithm. ey verified their algorithm on various 10-unit systems. e authors enhanced a novel technique [30], for thermal plants' dispatch generating powers based on motion optimization algorithm (IMA). ey achieved ELD being an objective function through implementing IMA. For the testing phase, multiple instances of various units of thermal plants were utilized to examine the execution of their algorithm.
eir preceding outcomes were matched with various approaches.
A detailed description of related ELD applications concerning different evolutionary approaches is given in Table 1.

Problem Formulation.
Emission can be included in economic dispatch's formulation in a variety of methods. Combined economic and emission dispatch (CEED) is one of the methods that are expressed as a problem of multiobjective optimization used to reduce emission and fuel costs to satisfy demand and avoid losses [38].

Combined Emission and Economy Dispatch (CEED).
CEED problem can be expressed as [39] N i�1 where P d represents total load demand, P l shows total transmission loss, and P i is the power produced by ith generator. Total fuel cost is denoted by FC, and total emission is denoted by EC. In (2), a i , b i , and c i represent fuel cost coefficients. From (3), the cost function depends on the problem nature, and it can be quadratic, square, sinusoidal, or any other function. Referred to (4), B ij coefficients or load flow can be utilized to find transmission losses denoted as P l , where B0 i Computational Intelligence and Neuroscience is the coefficient vector of B ij and a value B 00 . e price penalty factor is utilized to transform the problem of multiobjective optimization into a single-objective optimization problem as follows: Each plant or price penalty factor can be found for a specific demand as follows: (i) e ratio between the average fuel cost and the average emission of the maximum power capacity of that plant is obtained as follows: where (U i ) is the ith unit of plant capacity. (ii) Plants are arranged in ascending order based on the value of the price penalty factor. (iii) Each unit (U i ) maximum capacity is added one at a time, beginning from the lowest value of b i , unit until P i ≥ P d . (iv) At this point b i , linked with the last unit of the process, is the price penalty factor "b," Rs/Kg for the provided demand of the load.

Emission-Controlled Economic Dispatched (ECED).
Emission-controlled economic dispatch (ECED) is another method to reduce the economy related to a specific limit of emission concerning a specific demand. ECED problem's primary concern is to discover the cost-effective placement of plants while fulfilling the losses and demand and keeping the permissible limit of emission; FC needs to be reduced directed to (1), that is, power balance constraint and emission limit constraint expressed as Here, the system's total emission limit is denoted by E limit .

Bee Swarming.
is extraordinary insect is one of the most remarkable creatures since the old times. Honeybees have been the topic of scientific research. Moreover, multiple books and experiments have been carried out about honeybees; for instance, Ribbands published "Behavior and the Social Life of Honeybees" in 1953. "Anatomy of the Honey Bee" was written by Snodgrass in 1956, and omas D. Seeley wrote "the wisdom of hive" in 1995. e anatomy of a bee is shown in Figure 1. A process known as swarming is carried out to form new honeybee colonies: Table 1: Detailed description of related ELD applications concerning different evolutionary approaches.

[27]
Self-adaptable differential evolution algorithm integrating with multiple mutation strategies (ADE-MMS) 4 DE algorithms are tested on the ten ELD problems with diverse complexities 11 [28] Differential evolution the algorithm denoted as DEa-AR IEEE 57-bus system 12 [31] Modified crow search algorithm (MCSA) Five different well-known test systems 13 [36] Multi-objective multi-verse optimization algorithm 140 bus system 14 [24] Multi-objective economic and environmental dispatch problem (EEDP) Five generation systems 15 [37] Coyote optimization algorithm (COA) Power system consisting thermal generator 16 [29] Motion optimization algorithm (IMA) Several cases of different units of thermal plants (i) e old colony is left by queen bees among some workers and scout bees. (ii) A swarming cycle is shown in Figure 2. A collection of thousands to tens of thousands of bees make up a swarm. (iii) ey make a cluster around the queen in some branch or a tree, and twenty-fifty scouts are sent out to discover some new proper hives. (iv) Finally, under the supervision of the scout's bees, all other bees fly to the new hive.
Scout bees check a hive for various standards to meet, such as it must be wide enough to hold the entire swarm, and the entrance must be small and must be at the bottom of the hive, and must get a particular amount of heat from sunlight [40]. Processes of decision-making of scout bees are the source of inspiration. When they find various proper hives, they select the best among them. e source of communication between scouts is the movement of their wings and legs, which is called bee dance. e new hive is selected after the agreement of 80% of the scouts [40].
In terms of algorithms, every hive utilized by a scout demonstrates a feasible solution of an artificial search agent, while the fittest hive expresses a global optimum solution, which is represented in Table 2. e characteristics of the hive, like its size, size of entrance, and location of the entrance, can be viewed as a solution's fitness function. e process of collective decision-making of scouts is expressed as fitness weight (fw) in the algorithm.

Fitness-Dependent Optimizer.
e reproduction process of swarm bees is replaced by this algorithm. e major portion of the algorithm is obtained from hive exploring the process of scout bees from a pool of suitable options. e algorithm starts with the random initialization of the artificial scout population within the search space of X i (i � 1, 2, . . . n); the position of every scout expresses a recently recognized hive. Scout bees keep on finding the more suitable hive; once they find a better hive, they neglect the previous better hive; the same is the case with the algorithm. Whenever it discovers a new, more suitable solution than the earlier determined solution is neglected. If they cannot find any other better solution than the previous one, they will consider the current solution as the best solution.
A mechanism of fitness weight and random walk is used to randomly explore the landscape by artificial scouts in this algorithm. e following equation expresses the movement of artificial scout bees: where i denotes the current search agent, x denotes an artificial scout bee (search agent), pace denotes the direction and movement rate, and t denotes the current iteration of the artificial scout bee. Pace usually depends on fw, that is, fitness weight, whereas pace ′ s direction fully depends on a  Figure 2: Bee swarming process cycle [17]. Computational Intelligence and Neuroscience random mechanism. erefore, minimization problems' fw can be measured as e current best global solution's fitness function value is denoted by x * i,t fitness , current solution's fitness function value is denoted by x i,t fitness , and the weight factor is expressed as wf, which can have only 0 or 1 value and is used to control fw. If wf � 1, then it shows a low possibility of coverage and a high level of convergence. But if it is equal to 0, then it will not have any effect on equation (3.9), so it can be ignored, and if the variable is wf � 0, it will present us a more stable search. However, it reverses as the value of the fitness function entirely depends on the optimization problem. But, the value of fw must be in the range of [0, 1]; still, in some situations, when fw � 1, for instance, if the recent solution is the global best solution or global best solution and the recent solution are the same or hold similar fitness value. Furthermore, a possibility exists when fw � 0, if x * i,t fitness � 0. Lastly, it must bypass the chances to divide a number with 0 in case x i,t fitness , so it must follow the following rules: where a random number with a range of [−1, 1] is denoted by r. e random walk can be implemented in a variety of ways, but here Levy flight is selected as its good distribution curve offers stable movements [22]. According to FDO mathematical complexity: its time complexity for every iteration is O(p * n + p * CF); here, p denotes the size of the population, problem dimensions are denoted by n, and cost of the objective function is CF. Space complexity for every iteration is O(p * CF + p * pace); here, pace denotes the best previous paces stored. From this point, the time complexity of FDO is proportional to the number of iterations. However, space complexity will remain identical throughout the sequence of iterations. For the calculation of objective value, FDO owns a simple tool for calculations; it only calculates one random number and fitness weight for every agent [41]. Similarly, DA alignment, attraction, separation, some random values, and distraction are required to be calculated, while a majority of them are accumulative and the value of one depends on the value of others making the calculations complicated [42].

Single-Objective
Optimization-Based FDO. FDO with single-objective optimization problems (FDOSOOPs) starts with the initialization of artificial scouts on random locations of search landscape by utilizing lower and upper boundaries. For each iteration, it selects the global best solution, after that each artificial scout bee is computed by using (9). en, the value of fw is examined to decide whether it is 1 or 0 and if x i,t fitness � 0. pace is generated by utilizing (10). But, a random number denoted by r of range [−1, 1] will be generated if fw >0 and fw <1. For calculation of pace, (10) will be utilized if the value of r is less than 0 and value of fw will have a negative sign; however, for r ≥0, the pace will be calculated with the help of (10) and fw will have a positive sign. Random selection of signs for fw will ensure the random search of artificial bees in all directions.
In FDO, direction and size of pace are controlled by the randomization method; however, only the direction of pace is usually controlled by this method; in such situations, the pace's size depends on fw. Whenever scout bees find a new solution, it is compared with the current solution to determine whether it is better or not based on a fitness function. e earlier solution is neglected if the better latest solution is obtained. Similarly, if it is not better than the previous value of pace, it will be used by the scout bee to continue. On the other hand, if a better solution cannot be achieved by utilizing the previous value of pace, then the current solution will be continued by FDO to the next iteration. In FDO, whenever a solution is acquired, the value is saved for utilization in the next iteration. Two minor alterations are required for the implementation of FDO in maximization problems. (9) should be replaced by (11) as it is the inverse variant of (9): en, the criteria for the selection of the best solution must be altered. e statement ″ if(X t+1,i fitness < X t,i fitness)" needs to be replenished with ″ if(X t+1,i fitness > X t,i fitness)."

Multi-Objective
Optimization-Based FDO. FDO with multi-objective optimization problems (FDOMOOPs) begins with the initialization of artificial scouts into two-dimensional search space (X i , Y i ) . Each scout bee in the search space of (X i , Y i ) can be defined as X i (i � 1, 2, . . . n) and Y i (i � 1, 2, . . . n). en, the value of fw is examined to decide whether its 1 or 0 and if x i,t fitness � 0 or y i,t fitness � 0. In both cases, the pace can be generated as fw � |x i,t fitness /x * i,t fitness | − wf and fw � |y i,t fitness /y * i,t fitness | − wf.

Enhanced Method.
Multidimensional and multi-objective optimization algorithms tend to perform better for solving the linear and nonlinear constraint problems than the single-dimensional and single-objective optimization algorithms, especially in the case of ELD, when both the emission rate and fuel cost need to be minimized to approach the total loss and power demand. FDO is a multiobjective metaheuristic algorithm and, therefore, best suitable for solving constrained ELD problems. It can be implied to complex problems with nonlinear approximation. In this thesis, the authors have carried out FDO to solve the ELD problem by minimizing fuel cost, emission allocation, and transmission loss. Besides, the authors have employed a novel variant of FDO, which incorporates novel population initialization techniques and employed sine maps to select the weight factor for FDO dynamically.

Population Initialization.
e swarm or group of swarms needs to be fired randomly to obtain their initial fitness solution in the optimization process. e entire process is called population initialization. e most conventional method to assign an initial location to each individual is through a random number generator following the normal distribution. However, the major drawback to using the random number generator is premature convergence and abnormal exploration and exploitation. e random number generator developed a random number between the internal of 0 and 1. e probability of obtaining an optimal solution in the case of local minima is reduced when the initial locations are directed far from the solution, and each individual requires more steps and iterations to seek the entire solution.
e swarm can be stuck into premature convergence during the searching process and lead to poor exploration. Similarly, as opposed to this, the probability of obtaining a global solution in the case of global minima is diminished when the primary positions are delivered too near around the search space while the solution is out of search space; hence, each individual requires more steps and iterations to seek the entire solution and can be stuck into the premature convergence and leads to poor exploitation.

3.4.2.
Quasi-Random Sequence Initialization. Quasi-random is a distinction of n-rows that occupies n-dimensional search space. It is also called a low-disparity sequence. However, the usual standard quasi-random sequences and odd numbers all give consistently suitable sequences. ere is a significant distinction between these two patterns, aside from the standardized manner. An identical arbitrary generator on (0, 1) will deliver sequences, so every preliminary has a similar likelihood of producing a point on equivalent subintervals, for instance [(0, 1/2), (1, 1/ 2)]. In this manner, it is attainable for n preliminaries to inadvertently all extend in the top half of the range, while the (n + 1) points fall inside the other of the two parts with a likelihood of 1/2. While this is not the situation with the quasi-random sequences, the generated sequences are obliged by a low-inconsistency prerequisite that has a net impact on centers being created in a profoundly connected way. To avoid the premature convergence problem in FDO, the authors have carried out one of the quasi-random sequences called the Sobol sequence for the population's initialization.

Quasi-Random Sequence Initialization.
Sobol sequence is a low discrepancy sequence that was first enhanced by mathematicians in Russia in 1967 [43]. It mimics the random distribution by appropriating a base of two to shape progressively better uniform edges of the required interval and afterward reorder the directions in each measurement. Following are prime steps to generate the Sobol sequences S d : (iii) It is a notable pattern against each dimensional vector that for the whole to reach towards the indispensable points S d . Furthermore, the second great feature would be that the forecasts of x in the low range of the dimensioned face of S d cover most of the search area in terms of optimization. (iv) Subsequently, the comparable center of S d does not meet the criteria because in lower measurements, numerous focuses will be at a similar spot, in this way unnecessary for the vital estimation.
e comparison of FDO population initialization with random numbers following the Sobol distribution and the uniform distribution is presented in Figures 3 and 4 respectively. e following equation is used in the standard FDO for the swarm to select their initial locations to seek the entire optimal solution: In the enhanced FDO, the authors have selected the interval of [0, 1] for generating both sequences uniform and Sobol sequences in the process of FDO population initialization. e enhanced equation for initializing the swarm in FDO is presented as follows: where [0, 1] in (13) and (14) represents the standard limits of both generated sequences. e uniform random positions can be seen in Figure 3 with very random locations and illpatterned sequences, which may lead to poor exploitation. As compared to the uniform random, the Sobol sequence comes up with a well-patterned sequence in Figure 4, which may lead the swarm to converge maturely.

Enhanced Approach for Updating Weight Factor.
e weight factor is revealed as wf, which can have particularly 0 or 1 utility and is utilized to control wf. If wf belongs to absolute 1, then it confers a low probability of coverage and a high level of convergence. However, if the weight factor wf belongs to 0, then it will not have any influence on (15), so it can be ignored; if the variable is wf � 0, it will present us a Computational Intelligence and Neuroscience more stable search. However, it reverses as the value of the fitness function entirely depends on the optimization problem: However, in unusual circumstances, when the weight factor belongs to absolute one as shown in (16), for instance, if the new solution is the global best solution or local best solution and the new solution is identical or operates similar fitness value: wf should be balanced enough to control the exploitation and exploration for the controlled convergence rate when leads to absolute 0 wf and absolute 1 wf. For this, the authors produce a chaotic effect by using sine maps to the weight factor wf between the interval of [0, 1].

Chaotic Sine Map.
Chaotic maps produce uncontrolled groupings during the metaheuristic algorithm. e authors practiced the benefit of a chaotic sine pattern to update the weight factor [22]. e sine map is chaotic and used to produce a quarter effect between the interval of 0 and 1. When the weight factor becomes skewed towards 0, the sine wave covers the low balance and controls the low convergence rate. Similarly, when the weight factor becomes skewed towards 1, the sine wave covers the high balance and controls the high convergence rate. is phenomenon iteratively maintains the balance with optimal weight factor throughout the last epoch. A chaotic sine map can be defined as where 0 < m < 4 is the controlling factor. e author chooses m � 0.3 with the most optimal sequence. In terms of the weight factor, the equation becomes e enhanced variant of FDO utilized the following equation to update the fitness weight fw: e flowchart of the enhanced FDO along the ELD application is presented in Figure 5.

Dataset Overview.
e performance of the enhanced variant of FDO is evaluated through 24 units taken from the 18-unit system and 20-unit system with each of the 6-unit case study chunks by optimizing the fitness function enlisted in (1). e parameter sets used in the experiment for each unit are listed below.
Total number of units used in the experiment � 24, total power demand � 400, 700, number of iterations � 100, 200, population size � number of bee scouts � 50, and the beta coefficient used for 24 units according to each chunk of 6 units in the exploring capacity with a power demand of 400 MW and 700 MW are presented as follows: Detail of 24 units used to minimize the fuel cost, emission allocation, and transmission loss with 400 and 700 power load is presented in Table 2. In Table 2, Pmin and Pmax represent the lower and upper plant limits, respectively, whereas other parameters can be defined as a � (. e dataset used for the simulation consists of two chunks with 12 generating thermal units each. e first 12 generating units (1 to 12) are taken from Sys_18 U with all plant limitations and beta coefficients as represented in (20).
Similarly, the last 12 generating thermal units (13 to 24) are taken from Sys_20 U with all plant limitations and beta coefficients as described in (20). All thermal units employed for the empirical analysis are ramp-limits-free and do not endure in the prohibited zone for the smooth objective function. Comparison of simulation results on the ELD problem (FDO vs. enhanced FDO) with nonlinear optimization on 100 epochs with a power demand of 400 and 700 is presented in Tables 3 and 4 Certainly, for our problem statement, the accentuation is to distinguish, which epoch setting requires the most minimal fuel cost and transmission loss to discover arrangements of a specific worthy quality. Furthermore, the power demand is also analyzed in a roundabout way, to give in any event a complex reflection of the complexities of the various calculations considered in our relative examination. e obtained results certainly take the fact that the population initialization and optimal fitness factor of FDO make some impact on the global best of FDO as compared to the enhanced variant of FDO in terms of optimal allocation emission. e reason behind developing each thermal generation chunk with 6 units is to investigate the impact of greater dispersion on the total fuel cost and minimum error. To visualize the obtained results emission allocation results, a convergence comparison of FDO with the enhanced variant of FDO on the first 6 thermal units with 100 epochs and different power demands are illustrated in Figure 6.
Moreover e main reason behind performing the ANOVA test is to find the significant difference between the standard and enhanced FDO in terms of minimization. e additional reason to perform the ANOVA test is to determine which parameter delivers outcomes with critical contrasts, considering the target value accomplished by enhanced FDO from each run of the considerable number of tests performed. Graphical representation of one-way ANOVA test comparison (optimal allocation emission) of FDO with the enhanced variant of FDO on the 24 thermal units with (100, 200 epochs) and several power demands are illustrated in Figures 8 and 9, respectively.

Optimal Allocation Emission.
It is earlier mentioned that each chunk of thermal units is tested on 100 independent runs with 100 and 200 epochs considering two different combinations of power demand. e authors observed a significant improvement in the optimal power allocation generated by the enhanced FDO for the first 6 thermal units as compared to the conventional FDO (referred to Table 4). All 5 units' results obtained by enhanced FDO outperformed FDO except the 6th unit with 162.2504561 optimal emission allocation on 100 epochs and 400 power demand.
As contrasted to the first chunk of thermal units, the performance of the enhanced algorithm was observed less when optimizing emission allocation. It can be seen from Table 4 that only thermal units 7 and 12 gained better emission rates, which lead to greater divergence of the whole population. However, thermal units 8, 9, 10, and 11 show equal empirical performance for both FDO and enhanced FDO with a 44.3 emission rate.
In the case of the third chunk, thermal units 13 to 18, the emission allocation rate is not significantly improved using enhanced FDO instead of the standard FDO. e enhanced version obtained 18 with unit 13, 58 with units 14, 15, 16, 17, and 58 with unit 18, on 100 epochs and 400 power demand, which shows a slight improvement. is slight impact of enhanced FDO reveals the impact of robust population initialization on the ELD emission allocation. Lastly, the fourth chunk of the thermal unit from Table 4 exhibits outstanding results of the Enhanced algorithm on the entire parameter setting except for the last thermal unit with a 19.62533477 emission rate.
From Table 5, when power demand raised 400 to 700, the enhanced algorithm also improves the swarm convergence, and hence, the optimal fitness factor works here. is phenomenon shows the inverse divergence of the global best computed with the enhanced fitness factor, which leads to the emission allocation of the thermal units 3 and 6 from 68.72201216, 162.2504561 to 83.36202838, and 347.0304788, respectively. Similarly, premature convergence is highly tackled by the enhanced algorithm when seeing a significant decrease in the average optimal allocation for the first six thermal units. Tables 5 and 6 show similar convergence behavior of enhanced FDO compared to the FDO using 200 epochs with 400 and 700 power demand. However, the optimal fitness factor produces less impact than the effect produced when testing on 100 epochs. is can be due to the dimension reduction that occurs in higher generations.   Figure 5: Flowchart for the enhanced FDO algorithm along with ELD application.

Fuel Cost.
Fuel cost minimization on the same hydroenergy and power demand is a big issue when several units are mimicking in parallel. is can be optimized by considering the current fuel cost as the global best for each of the individuals in the FDO. However, the minimal risk is premature convergence, which leads to double computing cases and wastage of time with the cost approximately equal to the standard fuel cost. Robust population initialization decreases the chance of premature convergence. Hence, the enhanced FDO used optimal fitness factors in combination with the Sobol operator to minimize the fuel cost. It can also be observed in Tables 3-5 that the fuel cost difference between FDO and enhanced FDO is notable for all 24 thermal units on 100 and 200 epochs with a power demand of 400 and 700 sequentially. However, this significant minimal difference in cost can impact the whole unit generation cost. e trends for each chunk of the thermal unit from Tables 4-6 confirm the directly proportional relationship between the power demand and the fuel cost directly. e minimization range is constant between them, which explicates the strong divergence and influential fitness factor. Increasing power demand will lead the fuel cost to increase with a constant proportion of difference produced by FDO and enhanced FDO.

Transmission Loss.
Transmission error is essential for significant distance potential transmission, and it grows with an expansion in the measure of capacity to be dispatched. erefore, the utilization of inexhaustible force from the sustainable plants close to the heap focuses diminishes the transmission losses. Appropriating the sustainable power sources all through the working time frames as opposed to utilizing them during their accessible period will assist with diminishing both expense and the transmission loss.
Compared to the optimal emission allocation and fuel cost, the authors have received the best optimal transmission loss results. e enhanced algorithm FDO reduced the loss     FDO, which overcome it to 9.27E-04, 0.0011, 7.88E-04, and 8.23E-04 for four chunks of the thermal unit with 100 epochs and 700 power demand sequentially. Tables 6 and 7 explicitly formulate the same trends between standard and enhanced FDO with 200 significant differences in transmission error on 200 epochs and 400 and 700 power demand. Figure 6 confirms the optimal convergence comparison in the case of optimal allocation emission (FDO with the enhanced variant of FDO on the first six thermal units with 100 epochs and different power demands). Figures 7 and 8 dispense the clear-cut transmission loss difference. Units are presented on the X-axis, while transmission loss is enlisted on Y-axis.
To validate the obtained results, ANOVA statistical analysis for Figures 9 and 10 reinforces the best performance of the enhanced FDO algorithm and encourages the solution for other constraints as well. e box representation for enhanced FDO with 400 power demand is proved as an optimal solution with optimal chunk. Similarly, the interval plot representation for enhanced FDO with 700 power demand is determined as an optimal solution with an optimal chunk.

Conclusion
is research work introduced a variant of the FDO algorithm motivated by scout bees in the hive exploring the process of seeking food from a pool of suitable options. e enhanced variant is utilized to solve the economic load dispatch problem. FDO and its modified version are motivated to upgrade the minimization capability during weight optimization of economic load dispatch. Each individual of the scout bee is represented as output power generated through each thermal unit. e study deals with three types of constraints in this work: power balance capacity, transmission loss, and optimal emission allocation. In the beginning, the exploration executed by enhanced FDO is dependent on a simplistic fitness factor that delivers a less optimal solution by sticking into local minima and transforms some of its decision variables through their constraint  violation. After applying the Sobol operator for population initialization and chaotic sine map for the optimal fitness, redistribution power operators are connected. e enhanced operator ensures the feasibility of a probable solution that the thermal unit will take as an input and barely estimate the balance power constraint. Furthermore, the enhanced population initialization approach consolidates a quasirandom Sabol sequence to create the initial solution in the multidimensional search space. A regular 24-unit system is applied with diverse power demands for experimental evaluation. e experiential results acquired utilizing the enhanced variant of FDO confirm the superior performance in terms of low transmission loss, low fuel cost, and low emission allocation compared to the standard FDO. As a part of our future work, the authors are inspired by the hybridization of FDO with other metaheuristic algorithms such as BA, DE, and PSO. e authors aimed at taking the best qualities from BA as local search capability, DE as optimal mutation factor, and PSO as inertia weight and incorporating them in FDO to achieve the best results. Furthermore, the authors are also interested in the finetuning of FDO parameters in combination with ELD constraint and their hyperparameter tuning. Additionally, the hybridized version of FDO will be evaluated to investigate the influence of objective evaluations on dimension reduction.

ELD:
Economic load dispatch POZ: Prohibited operating zone VPE: Valve-point effects QBA: Quantum bat algorithm CEED: Combined economic emission dispatch ADE-MMS: Evolution algorithm integrating with multiple mutation strategies WEV: EVs and wind farms WE_DEED: Dynamic economic emission dispatching based on WEV system IMA: Motion optimization algorithm BA: Bat algorithm ABC: Artificial bee colony CSA: Chaotic-based self-adaptive NSGA-RL: Learner nondominated sorting genetic algorithm CCDE: Chaotic-crisscross differential evolution DEA: Differential evolution algorithm ADE-MMS: Self-adaptable differential evolution algorithm integrating with multiple mutation strategies MCSA: Modified crow search algorithm EEDP: Multi-objective economic and environmental dispatch problem COA: Coyote optimization algorithm CEED: Combined economic and emission dispatch ECED: Emission-controlled economic dispatch FDO: Fitness-dependent optimizer FDOSOOPs: FDO with single-objective optimization problems FDOMOOPs: Multi-objective optimization problems.

Data Availability
Data can be shared upon request from the corresponding author.

Conflicts of Interest
e authors declare that they have no conflicts of interest. 14 Computational Intelligence and Neuroscience