Real-time adaptive load frequency control in hybrid microgrids using crayfish optimization and balloon effect: A path toward diesel-free sustainable operation

Abstract

The growing penetration of renewable energy sources (RESs) in modern microgrid architectures has intensified the need for real-time, resilient, and self-adaptive load frequency control (LFC) systems—particularly under low-inertia conditions and highly intermittent generation. This paper proposes a novel real-time adaptive LFC framework for standalone PV–HFC hybrid microgrids, designed to operate with near-zero reliance on diesel backup. The proposed strategy integrates a disturbance-sensitive Balloon Effect (BE) identifier for online sensitivity estimation with the Crayfish Optimization Algorithm (CrOA) for continuous gain adaptation of a structurally simple integral controller. Extensive simulation experiments under severe operating scenarios—including sudden load disturbances, parametric uncertainty, and renewable source disconnection—demonstrate that the CrOA+BE controller significantly outperforms benchmark metaheuristics such as GTO, Jaya, and SCA. The framework is further validated through hardware-in-the-loop (HIL) implementation, ensuring practical readiness for real-time deployment. Remarkably, the system achieves up to 95% reduction in diesel consumption, leading to major savings in operating cost and CO₂ emissions. Overall, the proposed scheme represents a robust pathway toward intelligent, fossil-free, and economically viable frequency regulation for next-generation renewable microgrids. The proposed controller explicitly integrates Information Technology through intelligent computational algorithms that enhance real-time microgrid stability.

Keywords

Hybrid microgrid Diesel generator Hydrogen Cell Crayfish Optimization real-time simulation renewable energy optimization algorithms

Introduction

Background

Rising global energy needs, combined with the adverse environmental consequences of fossil-fueled power generation, have prompted a strong international push toward renewable and low-emission energy sources. Despite the dominance of coal, oil, and natural gas in electricity production across many regions, these fuels are responsible for the majority of greenhouse gas (GHG) emissions—contributing roughly 73% of global CO₂ emissions.¹ Moreover, thermal power plants are not only carbon-intensive but also water-intensive, consuming vast quantities of freshwater and emitting harmful pollutants such as nitrogen oxides and particulates.^2,3 Conversely, renewable energy (RE) options like solar photovoltaics (PV), wind, and hydrogen derived from clean sources have gained considerable momentum due to their zero-emission profiles, widespread availability, and cost competitiveness.^4–6 Solar PV, in particular, has seen rapid deployment, with over 270 million tons of CO₂ avoided in 2021 alone,⁷ and evidence suggests that it can reduce air pollution in cities by more than 30%.^8,9 Additionally, because PV systems do not require water during operation, they present a viable solution in regions facing water scarcity.^10,11

However, incorporating RE technologies (especially in isolated microgrids (µGs)) poses technical difficulties due to their variability and lack of inertia.^12,13 These systems often operate without the stabilizing presence of conventional synchronous machines, rendering them vulnerable to frequency instabilities during supply-demand mismatches.^14,15 To maintain grid stability in such environments, flexible energy storage systems are essential. Hydrogen fuel cells (HFCs) are among the leading candidates, offering rapid response capabilities and full alignment with decarbonization goals.^16,17 When integrated with battery storage, HFCs can support effective frequency control, reliable energy buffering, and long-term storage functionality.^18–20

Motivation

Although renewable energy-based microgrids have demonstrated strong potential for sustainable electrification, especially in remote and underserved regions,^21,22 their practical deployments still depend heavily on diesel generators or other fossil-fuel backup units to maintain frequency stability under uncertain and intermittent power conditions.^23,24 This dependency significantly undermines the environmental and economic benefits of renewables, resulting in continued carbon emissions, high fuel logistics cost, and frequent maintenance requirements.^25–27 Furthermore, most existing LFC strategies either rely on fixed-gain PI/I controllers or employ offline-tuned metaheuristic algorithms, which cannot adapt to fast-changing disturbances, communication delays, or nonlinear interactions in highly dynamic hybrid microgrids.^28–33 Even recent optimization-based LFC approaches still fall short in real-time adaptability and lack experimental (HIL) validation, raising concerns about their practical deplorability.

Meanwhile, Hydrogen Fuel Cells (HFCs) have emerged as a promising zero-emission and fast-response alternative to diesel, but their full potential has not yet been realized in intelligent real-time LFC frameworks.^34–36 Likewise, the Crayfish Optimization Algorithm (CrOA), despite its strong optimization capability, has not yet been applied to real-time adaptive frequency regulation in renewable-dominant microgrids.^37–40 This critical gap (the absence of a real-time, fossil-free, experimentally validated adaptive LFC solution) directly motivates the present research. Accordingly, this work proposes a CrOA+BE–driven online adaptive control framework that enables resilient, low-emission, and real-time frequency stabilization, aligning with global sustainability goals (e.g., SDGs 7, 9, and 13).

State of the art

As the complexity of modern microgrids continues to grow (particularly those dominated by renewable energy sources (RESs)) the demand for more advanced and responsive load frequency control (LFC) strategies has intensified.^41,42 While conventional controllers such as proportional-integral (PI) and integral-only types remain widely used due to their ease of implementation, their fixed-parameter configurations are ill-suited for the dynamic behavior introduced by renewable variability and irregular load profiles.⁴³ In practice, these traditional schemes tend to struggle under conditions involving nonlinear dynamics, communication lags, and unpredictable disturbances.⁴⁴ To address these shortcomings, many researchers have adopted metaheuristic optimization techniques that support real-time, self-adjusting control strategies.⁴⁵ Algorithms like Grey Wolf Optimization (GWO), Jaya, Whale Optimization Algorithm (WOA), Harris Hawks Optimization (HHO), and Artificial Gorilla Troops Optimization (GTO) have all demonstrated their effectiveness in enhancing system performance—offering quicker response, better disturbance rejection, and lower overshoot.⁴⁶ Still, these methods are not without drawbacks: GWO can be slow to converge, Jaya tends to falter in high-noise environments, and both HHO and WOA require considerable computational overhead, which may hinder practical deployment due to premature convergence issues.⁴⁷

Despite these developments, many existing control approaches continue to depend on diesel generators as backup, which ontradicts the sustainability goals often associated with renewable integration.⁴⁸ Hydrogen Fuel Cells (HFCs), though cleaner and capable of fast dynamic response, remain underutilized in adaptive frequency regulation schemes.⁴⁹ Similarly, the Crayfish Optimization Algorithm (CrOA)—a relatively recent bio-inspired method with strong balance between exploration and exploitation—has seen limited application in LFC contexts.⁵⁰ The Balloon Effect (BE) identifier, which offers potential for enhancing controller responsiveness to abrupt changes, has also received little attention.⁵¹ Furthermore, a key limitation across much of the literature is the reliance on purely simulated validation, with minimal engagement in hardware-in-the-loop (HIL) testing or experimental trials.⁵²

In recent studies, researchers have begun incorporating adaptive mechanisms such as the Balloon Effect (BE) into metaheuristic-tuned controllers to enhance transient awareness and sensitivity. For example, SCO+BE–based controllers in ⁴⁰ demonstrated faster response during abrupt disturbances, but were applied only to motion control systems rather than real microgrids, and lacked any discussion of fossil-free operation or real-time deployment. Likewise, GTO+BE strategies applied to PV-fed microgrids⁵³ showed improved damping and emission reduction, but still retained diesel support, and were validated only through offline simulation—not HIL. BE-supported Bat Algorithm (BA+BE) schemes have also emerged, yet these rely on batch/offline tuning rather than online adaptive learning, making them less suited for time-critical LFC applications.⁴⁵ Additionally, several HFC-based fuzzy or fractional controllers have been proposed in 2023–2024 literature, but they do not integrate intelligent online gain tuning, nor do they achieve fully diesel-free operation in practice. Despite these advancements, no existing study provides a real-time, fossil-free, CrOA+BE-based adaptive LFC with HIL validation—which directly motivates the present work.

Research gap

Although significant progress has been made in developing frequency regulation techniques for microgrids powered by renewable energy, several critical research gaps remain unresolved. One of the most persistent issues is the continued dependence on fossil-fueled backup systems, particularly diesel generators. This reliance underscores the technical and operational challenges that hinder a complete transition to renewable-only configurations. Diesel generators still serve as essential components for ensuring grid stability; however, their negative environmental footprint highlights the urgency of integrating cleaner alternatives to reduce fossil fuel reliance.

Hydrogen Fuel Cells (HFCs) offer an attractive solution due to their quick dynamic response, modularity, and emission-free nature. Despite these advantages, their incorporation into adaptive control strategies has seen limited development. On the algorithmic front, although established metaheuristic methods are frequently used for optimizing controller parameters, newer approaches—such as the Crayfish Optimization Algorithm (CrOA), known for its strong global and local search capabilities—have not been adequately explored for load frequency control (LFC). In parallel, techniques like the Balloon Effect (BE) Identifier, which can significantly improve control sensitivity under rapid or irregular system changes, remain underutilized.

Another major shortcoming in the current body of research lies in how these control methods are evaluated. Most studies are restricted to simulation environments and lack real-time validation through platforms such as hardware-in-the-loop (HIL). As presented in Table 1, while theoretical improvements are commonly reported, practical implementation issues in operational hybrid microgrids are rarely addressed. To close these gaps, the present study introduces a real-time adaptive control framework for a hybrid microgrid that combines renewable sources, hydrogen fuel cells, and diesel generation. The proposed system aims to deliver enhanced performance, reduce environmental impact, and maintain operational dependability—advancing the broader goal of sustainable energy development. Table 1 clearly exposes a critical gap—none of the existing works deliver a fully real-time, fossil-free, Crayfish Optimization+Balloon Effect–driven LFC solution with verified HIL implementation. This is precisely the gap addressed in the present study.

Table 1.

Comparative summary of previous control strategies in hybrid microgrids.

No.	Reference	Control/Tuning	Storage/Plant	HIL	Fossil-free	Key takeaway
1	⁵⁴	PI+HHO	Thermal/Hybrid	No	No	Improved tuning but no real-time or HIL
2	⁵⁵	PID+SCA	RES-based µG	No	No	Chaos-based tuning, no HIL or fossil-free
3	⁵⁶	PID+Jaya	4 area system	No	No	Offline tuning only
4	⁵⁷	LFC+GTO	PV+thermal	No	No	Strong simulation, still fossil+no HIL
5	⁵⁸	Integral+SCO+BE	EV/HPWH µG	No	No	Adaptive via BE, no HIL or HFC
6	⁵⁹	Fuzzy/FO+HFC	PV+HFC	No	Partial	HFC-based but no real-time CrOA+BE
7	⁶⁰	HIL evaluation (benchmark)	GFM µG	Yes	—	Validation only, not a CrOA+BE controller
8	⁴⁰	SCO+BE	Motion/Position control	NO	NO	Not LFC, no HFC, not fossil-free, no real-time or HIL validation
9	⁵³	GTO+BE	PV microgrid	No	No	Still diesel-dependent, simulation-only, no HFC, no HIL validation
10	This work	Integral+CrOA+BE (online)	PV+HFC	Yes	Yes	First real-time fossil-free CrOA+BE with HIL

Research objectives and contributions

This study aims to develop a real-time adaptive load frequency control (LFC) framework for standalone hybrid microgrids that integrate photovoltaic (PV) generation with Hydrogen Fuel Cells (HFCs) as the primary clean energy storage and near-zero reliance on diesel backup. The objective is to ensure fast, robust, and sustainable frequency regulation under realistic operating uncertainties such as abrupt load disturbances, renewable intermittency, and parameter variations.

The proposed control strategy is built upon a structurally simple integral controller that is enhanced with two intelligent real-time mechanisms.

➢ Balloon Effect (BE) Identifier—provides online disturbance sensitivity estimation to detect system stress in real time.

➢ Crayfish Optimization Algorithm (CrOA)—performs adaptive online gain tuning to guarantee optimal response under varying conditions.

This CrOA+BE synergy enables self-learning, disturbance-aware, and high-resilience frequency control, surpassing the limitations of offline-tuned or fixed-parameter controllers.

Main Contributions and Innovations of This Work.

Unlike previous LFC studies, this work introduces the first real-time CrOA+BE-driven adaptive control architecture that is designed for fossil-free operation (when diesel is disabled) and is experimentally validated through hardware-in-the-loop (HIL)—a combination not previously reported in existing literature. To validate the effectiveness of the proposed approach, a comparative performance analysis will be conducted between CrOA and other well-established metaheuristic algorithms, including Artificial Gorilla Troops Optimization (GTO), Jaya, and Sine Cosine Algorithm (SCA).

The key scientific contributions are summarized as follows:

• First application of CrOA for real-time adaptive frequency control in renewable-dominant microgrids.

• Novel integration of BE-based online disturbance sensitivity with CrOA for self-adjusting gain tuning, ensuring superior robustness against abrupt grid fluctuations.

• Validated through both simulation and HIL, bridging the gap between theoretical development and practical field deployment.

• Achieves up to 95% reduction in diesel dependency, significantly lowering fuel cost and emissions.

• Includes an economic evaluation (LCOE+payback period), confirming its financial viability for real-world adoption.

• Aligns with global sustainability targets, directly supporting SDG 7 (Clean Energy), SDG 9 (Innovation), and SDG 13 (Climate Action).

Power system model

The microgrid analyzed in this research, depicted in Figure 1, functions independently from the main grid in islanded mode. It consists of three primary elements: a 30 MW diesel generator that supplies dispatchable power, a 6 MW solar photovoltaic (PV) array contributing variable renewable energy, and a Hydrogen Fuel Cell (HFC) that operates under dynamic control to store energy and assist in maintaining frequency stability. The system is designed to support a constant base load demand of 18.6 MW, which is classified as critical and cannot be adjusted. To ensure frequency stability, a centralized control system coordinates the responses of both the diesel generator and the HFC.

Figure 1.

Hybrid microgrid architecture with diesel generator, PV, and HFC units.

To model the behavior of the microgrid, a state-space framework is used, which captures the evolution of internal states over time, incorporating the effects of control actions and external influences

\dot{X} = AX + BU

(1)

Y = CX + DU

(2)

In these formulations, X refers to the internal state vector of the system, U encompasses both control actions and external disturbances, and Y signifies the system’s output, typically focusing on frequency deviation. The system’s natural behavior is captured by the matrix A, while matrices B and D illustrate how external inputs affect the progression of internal states and the resulting outputs, respectively

\dot{X} = [\begin{array}{c} \frac{- D}{2 H} & 0 & \frac{1}{2 H} & \frac{1}{2 H} \frac{1}{2 H} \\ \frac{- 1}{R T_{g}} & \frac{- 1}{T_{g}} & 0 & 0 0 \\ 0 & \frac{1}{T_{t}} & \frac{- 1}{T_{t}} & 0 0 \\ 0 & 0 & 0 & \frac{- 1}{T_{P V}} \frac{- 1}{T_{H F C}} \end{array}] \times [\begin{array}{c} Δ f \\ Δ P_{g} \\ Δ P_{d} \\ Δ P_{P V} \\ Δ P_{H F C} \end{array}] + [\begin{array}{c} 0 \\ \frac{- 1}{T_{g}} \\ 0 \\ 0 \\ \frac{- K_{H F C}}{T_{P V}} \end{array}] \times [Δ P_{c}] + [\begin{array}{c} 0 & \frac{1}{2 H} \\ 0 & 0 \\ 0 & 0 \\ \frac{1}{T_{P V}} & 0 \end{array}] \times [\begin{array}{l} Δ P_{solar} \\ Δ P_{L} \end{array}]

(3)

Y = [\begin{array}{l} 1 & 0 & 0 & 0 \end{array} 0] \times [\begin{array}{c} Δ f \\ Δ P_{g} \\ Δ P_{d} \\ Δ P_{P V} \\ Δ P_{H F C} \end{array}] + [0] U

(4)

A basic representation of frequency behavior, derived from the mismatch between power supply and demand, can be formulated as

Δ f = (\frac{1}{M}) . (Δ P d + Δ P_{H F C} - Δ P_{L} - Δ P_{P V}) - (\frac{D}{M}) . Δ f

(5)

In this context, Δf denotes the deviation in system frequency, ΔPd represents the variation in power output from the diesel generator, and ΔPL indicates the change in load demand. The parameter M stands for the system’s equivalent inertia, while D corresponds to the damping coefficient.

The time-dependent power output of the diesel generator (DGP) can be described by the following expression

Δ P d = (\frac{1}{T d}) . Δ P g - (\frac{1}{T d}) . Δ P d

(6)

likewise, the governor’s (GD) power response is expressed as

Δ P g = (\frac{1}{T g}) . Δ P c - (\frac{1}{R . T d}) . Δ f - (\frac{1}{T g}) . Δ P g

(7)

In this formulation, ΔPg refers to the change in the governor’s output, while ΔPc represents the additional control input. The parameters R, Tg, and Td correspond to the governor droop, the turbine’s time constant, and the governor’s time constant, respectively. Together, these equations capture how the system responds to disturbances affecting frequency, and they highlight the role of control mechanisms (especially those involving the diesel generator and the hybrid frequency controller (HFC)) in maintaining or restoring frequency stability.

Hydrogen fuel cell modeling

The hydrogen fuel cell (HFC) is incorporated into the frequency control system to mitigate sudden changes in power demand or supply variations. Its dynamic response is represented by a first-order transfer function as follows

G_{F C} (S) = \frac{K_{F C}}{{1 + S . T}_{F C}}

(8)

{Δ P}_{H F C} = \frac{K_{H F C}}{{1 + S . T}_{H F C}} . Δ P c

(9)

where

K_{F C}

denotes the gain of the HFC system, and

T_{F C}

represents its time constant. This model describes how the HFC power output gradually responds to the control signal. The generated power

{Δ P}_{H F C}

is then injected into the system to assist in maintaining frequency stability alongside the diesel generator. This modeling approach aligns with the method presented in¹⁸, which demonstrates the effectiveness of using first-order approximations to simulate HFC dynamic behavior for frequency regulation. Additional studies, such as,^18–20 further support the accuracy and feasibility of this model in real-time microgrid control applications.

Crayfish Optimization Algorithm (COA)

The Crayfish Optimization Algorithm (COA), introduced by Heming and colleagues, is based on the behavioral patterns of crayfish (specifically foraging), competition, and seeking summer refuges. These behaviors are mapped into three algorithmic phases that balance exploration and exploitation. The summer resort phase represents exploration, while competition and foraging relate to exploitation. At the start, a population of solutions (referred to as a crayfish colony) is initialized and represented as $= X_{1}, X_{2}, X_{3}, \dots \dots \dots ., X_{\dim}$ , where each $X_{i}$ denotes the position of an individual solution in a multi-dimensional search space. Random factors simulating ambient temperature influence crayfish actions. As temperature rises, the algorithm transitions into the summer resort or competition phase. In the summer resort stage, solutions are adjusted based on individual positions and shaded locations ( $X_{s h a d e}$ ). If conditions are favorable, the algorithm proceeds to the foraging phase, where the best solution (viewed as a food source) is targeted.

During foraging, each crayfish evaluates its current solution using a fitness function and compares it with an ideal solution, adjusting its behavior based on the size and quality of the “food.” If the food is too large, crayfish use their limbs to break it down before proceeding. Overall, COA mimics crayfish survival strategies by alternating between exploration (temperature-based) and exploitation (solution refinement), seeking optimal solutions through this biologically inspired process.⁶¹

Initialization phase of the CrOA

In this research, a controller is introduced that incorporates both minimum and maximum bounds for its parameters. To guide the algorithm toward achieving a global objective, the initial population in the Crayfish Optimization Algorithm (CrOA) is generated within these defined limits using fundamental mathematical operations. The initialization of each individual in the population is carried out using the following equation

X_{i, j} = rand \times (U B_{j} - L B_{j}) + L B_{j}

(10)

where

rand

is a random number between 0 and 1, and

U B_{j}

and

L B_{j}

represent the upper and lower bounds for the

j^{t h}

parameter, respectively.

The CrOA algorithm begins by randomly creating a set of candidate solutions, denoted as $X$ . During each iteration, the solution that comes closest to the global target is identified as the current optimal solution. The positions of all candidate solutions are organized and presented in the following matrix

X = [\begin{array}{l} X_{1, 1} & X_{1, j} & \dots & X_{1, \dim} \\ X_{i, 1} & X_{i, j} & \dots & X_{i, \dim} \\ ⋮ & ⋮ & ⋱ & ⋮ \\ X_{N, 1} & X_{N, j} & \dots & X_{N, \dim} \end{array}]

(11)

in this context,

X

represents the initial positions of the population,

N

denotes the total number of individuals in the population, and

\dim

refers to the number of dimensions in the search space.

The fitness of each solution is evaluated, and the one with the highest fitness value among all candidates is identified as the best solution. The calculation of the fitness value is carried out using the following expression

f_{fitness} = {[f 1 f 2 f 3 \dots \dots \dots \dots f_{N}]}^{T}

(12)

Crayfish temperature and food intake definition

The behavior of crayfish is heavily influenced by temperature variations, which dictate the phase they enter within the algorithm. The temperature is calculated according to equation (13). When the ambient temperature rises above 30°C, crayfish instinctively seek cooler environments, corresponding to the summer resort phase. When temperatures are within a favorable range, crayfish shift to the foraging phase.

Temperature not only affects their behavioral phase but also governs their food intake. Crayfish exhibit optimal feeding behavior when the temperature ranges between 15°C and 30°C, with 25°C being the most ideal. The relationship between temperature and crayfish intake is mathematically modeled using equation (14)

T e m p = r a n d * 15 + 20

(13)

In this equation, “

T e m p

” represents the environmental temperature surrounding the crayfish

p = C_{1} \times (\frac{1}{\sqrt{2 \times π} \times σ} \times \exp (\frac{- {(Tem p - μ)}^{2}}{2 σ^{2}}))

(14)

Among these parameters, σ and C₁ are employed to adjust the crayfish’s food intake based on varying temperatures, while µ indicates the optimal temperature preferred by crayfish.

Exploration phase (summer resort stage)

When the temperature rises above 30°C, it is deemed excessively high, prompting crayfish to seek shelter in caves as part of their summer vacation behavior. The location of the cave, denoted as $X_{s h a d e}$ , is determined using the following expression

X_{s h a d e} = (X_{G} + X_{L}) / 2

(15)

In this context,

X_{G}

represents the best position achieved so far across all iterations, while

X_{L}

denotes the best position within the current population.

The competition among crayfish for cave spaces occurs randomly. If a randomly generated number satisfies $r a n d < 0.5$ , it indicates that no other crayfish are competing for the same cave. In such cases, the crayfish move directly to the identified cave to take shelter during their summer break. This behavior is mathematically described by equation (16)

X_{i, j}^{t + 1} = X_{i, j}^{t} + C_{2} \times rand \times (X_{shade} - X_{i, j}^{t})

(16)

Here,

t

denotes the current iteration number, while

t + 1

represents the iteration number of the next generation. The parameter

C_{2}

corresponds to the decreasing curve, as defined by equation (17)

C_{2} = 2 - (t / T)

(17)

In this case,

t

represents the total number of maximum iterations.

During the summer resort phase, the primary objective of the crayfish is to approach the cave, symbolizing the optimal solution. As a result, crayfish move toward the cave, striving to minimize their distance to the best solution. Through this movement, individuals progressively get closer to the optimal point, enhancing the exploitation ability of the CrOA algorithm and leading to faster convergence toward the global optimum.

Exploitation phase (competition stage)

When the temperature surpasses 30°C and the randomly generated value satisfies $r a n d \geq 0.5$ , it suggests that multiple crayfish are attracted to the same cave. In such cases, a competitive battle ensues among the crayfish to claim access to the cave. This competition behavior is mathematically described by equation (18)

X_{i, j}^{t + 1} = X_{i, j}^{t} - X_{z, j}^{t} + X_{shade}

(18)

Here,

z

represents the randomly selected individual crayfish, as defined in equation (19)

z = r o u n d (r a n d \times (N - 1) + 1)

(19)

During the competition stage, crayfish compete for access to caves. Each crayfish, represented by $X_{i}$ , updates its position relative to another randomly selected crayfish, $X_{z}$ , as expressed in equation (18). This dynamic adjustment broadens the CrOA algorithm’s search space, thereby strengthening its exploration capability.

Exploitation phase (foraging stage)

When the temperature $T e m p \leq 30$ , conditions are favorable for crayfish to feed. Under these circumstances, crayfish continuously move toward the food source. Upon reaching the food, they evaluate its size. If the food is found to be too large, crayfish use their claws to tear it into smaller pieces. They then consume the food using their two remaining walking legs in an alternating manner. The location of the food source, denoted as $X_{f o o d}$ , is determined by the following expression

X_{f o o d} = X_{G}

(20)

The size of the food, denoted by $Q$ , is defined as follows

Q = C_{3} \times r a n d \times ({f i t n e s s}_{i} / {f i t n e s s}_{f o o d})

(21)

C_{3}

represents the food factor, indicating the maximum food size, and it is assigned a constant value of 3. The term

{f i t n e s s}_{i}

corresponds to the fitness value of the

ith

crayfish, while

{f i t n e s s}_{f o o d}

refers to the fitness value at the food location. The crayfish use the maximum food quantity as a reference point to determine the size of the food.

When $Q$ exceeds $(C_{3} + 1) / 2$ , it indicates that the food quantity is considered large. In such a case, the crayfish will employ their primary claw to tear apart the food. The mathematical expression describing this behavior is given as follows

X_{food} = \exp (\frac{- 1}{Q}) \times X_{food}

(22)

This mechanism allows crayfish to adjust their feeding strategy based on the size of the available food, promoting efficient consumption and optimal resource use.

Once the food is broken into smaller, crumbly pieces, crayfish will alternately use their second and third walking legs to pick up the food and bring it to their mouths. This alternating motion is modeled using a combination of sine and cosine functions. Additionally, the amount of food intake is directly linked to the quantity of food obtained. The foraging behavior is mathematically described as follows

X_{i, j}^{t + 1} = X_{i, j}^{t} + X_{food} \times p \times (\cos (2 \times π \times rand) - \sin (2 \times π \times rand)

(23)

When

Q \leq (C_{3} + 1) / 2

, the crayfish simply moves directly toward the food and consumes it immediately. The mathematical expression describing this behavior is given below

X_{i, j}^{t + 1} = (X_{i, j}^{t} - X_{food}) \times p + p \times rand \times X_{i, j}^{t}

(24)

During the foraging phase, crayfish adopt various feeding strategies based on the size of the food, represented by $Q$ . In this context, the food location $X_{food}$ symbolizes the optimal solution. When the size $Q$ is suitable for feeding, the crayfish simply move toward the food and consume it. However, if $Q$ is too large, it indicates a considerable gap between the crayfish’s current position and the optimal solution. In such situations, $X_{food}$ must be adjusted to minimize this gap and move the crayfish closer to the food source.

Additionally, the randomness incorporated into the crayfish’s food intake adjustment is managed under controlled execution, ensuring a more effective movement toward the optimal solution during the foraging process. Through these adaptive behaviors, the CrOA algorithm progressively moves closer to the optimal point, thereby enhancing its exploitation ability and improving its convergence performance. The overall process is illustrated in the flowchart shown in Figure 2. While Figure 3 illustrates the proposed control loop, it shows how CrOA adaptively tunes the PI controller to regulate the HFC unit, ensuring fast correction of frequency deviations.

Figure 2.

Flowchart of the crayfish optimization algorithm (CrOA)⁶¹.

Figure 3.

Schematic of crayfish behavior phases in CrOA.

Adaptive frequency control

BE identifier

The balloon effect (BE) refers to the influence of air pressure on the size of a balloon. In a similar manner, disturbances and parameter uncertainties in a system can significantly impact $G_{i} (s)$ . Figure 4 illustrates how the BE identifier influences the objective function of an optimization strategy during each iteration. By incorporating this technique, the overall algorithmic process is improved, as demonstrated in studies.⁴⁰

Figure 4.

Effect of balloon effect (BE) on objective function in optimization process.

The transfer function representing the studied microgrid at any given moment $i$ is expressed as follows

G_{i} (s) = \frac{Y_{i} (s)}{U_{i} (s)}

(25)

Additionally,

G_{i} (s)

is dependent on its previous value,

G_{i - 1} (s)

. Here,

{A L}_{i}

denotes a gain factor, while

G_{0} (s)

represents the nominal transfer function of the process

G_{i} (s) = {A L}_{i} G_{i - 1} (s)

(26)

G_{i - 1} (s) = ρ_{i} G_{0} (s)

(27)

where

ρ_{i} = \prod_{n = 1}^{i - 1} {A L}_{n}

(28)

G_{i} (s) = {{A L}_{i} ρ}_{i} G_{0} (s)

(29)

In the proposed framework, the Balloon Effect (BE) acts as an online disturbance sensitivity estimator. At each control cycle, it continuously monitors the instantaneous frequency deviation and evaluates the corresponding control effort injected toward the load interface. Accordingly, the adaptive load sensitivity factor (AL_i) is computed in real time as the ratio between the observed frequency response and the applied control action

{A L}_{i} = \frac{Δ f (t)}{U_{i} (s)}

(30)

where Δf(t) represents the instantaneous deviation from nominal system frequency, and u(t) denotes the control signal before being applied to the load. This direct coupling allows the BE to quantify the severity of the system stress and trigger proportional adjustment of the controller gain. By doing so, the BE ensures fast transient awareness and adaptive reinforcement of control aggressiveness during abrupt load/renewable disturbances—without requiring any offline tuning.

In summary, the balloon effect (BE) functions as an online identifier capable of detecting system challenges, such as load disturbances or parameter variations. This capability enhances the performance of the integrated optimization technique.

CrOA+BE

Figure 5 presents a simplified model of the microgrid, which is used to determine the parameters of a second-order closed-loop system for the controlled area

T . F = \frac{(\frac{k_{i} A L_{i} ρ_{i}}{M_{0}})}{s^{2} + (\frac{D_{0} + A L_{i} ρ_{i}}{M_{0}}) s + (\frac{k_{i} A L_{i} ρ_{i}}{M_{0}})}

(31)

ω_{n, i} = \sqrt{(\frac{k_{i} A L_{i} ρ_{i}}{M_{0}})}

(32)

η_{i} = \frac{(\frac{D_{0} + A L_{i} ρ_{i}}{M_{0}})}{2 ω_{n, i}}

(33)

M_{p, i} = e^{(\frac{- π (D_{0} + A L_{i} ρ_{i})}{2 M_{0} ω_{n, i} \sqrt{1 - η_{i}^{2}}})}

(34)

T_{r, i} = \frac{π - \sqrt{1 - η_{i}^{2}}}{ω_{n, i} \sqrt{1 - η_{i}^{2}}}

(35)

T_{s, i} = \frac{8}{(\frac{D_{0} + A L_{i} ρ_{i}}{M_{0}})}

(36)

the OF of the CrOA+BE identifier

J = \min \sum {(T_{r, i}^{2} + T_{s, i}^{2} + M_{p, i}^{2})}^{0.5}

Figure 5.

Simplified model for controlled microgrid area.

Results and discussion

This section presents the evaluation of the proposed adaptive load frequency control (LFC) strategy based on the Crayfish Optimization Algorithm (CrOA) integrated with the Balloon Effect (BE) identifier. To ensure fair and consistent comparison, all benchmark optimization techniques (namely Jaya, Gorilla Troops Optimizer (GTO), and Sine Cosine Algorithm (SCA)) are also combined with the BE mechanism. The hybrid microgrid under investigation, illustrated in Figure 1, consists of a diesel generator, a hydrogen fuel cell (HFC), and a photovoltaic (PV) array operating in an islanded mode. The system performance is assessed under five distinct scenarios: step load increase, system parameter variation, variable PV generation, sudden disconnection of the HFC, and temporary reduction in HFC capacity. For each scenario, the system’s dynamic response is evaluated and compared in terms of frequency deviation, maximum overshoot, settling time, and performance indices such as the Integral of Squared Error (ISE) and Integral of Time-weighted Absolute Error (ITAE), in order to highlight the effectiveness of the CrOA+BE strategy relative to other optimization methods under identical enhancement conditions. All simulations are executed on a high-performance desktop PC equipped with Windows 10 operating system, Intel Core i5 processor (4.2 GHz), and 16 GB RAM, utilizing MATLAB R2014 software supported by the QUARC sub-program and the QUARC pid_e data acquisition card. The nominal parameters of the system, along with the optimization algorithms parameters, are provided in Tables 2 and 3. It is important to note that no fixed or manually-tuned parameters are required. All controller gains are adaptively adjusted online using the Crayfish Optimization Algorithm (CrOA), while the Balloon Effect (BE) identifier continuously detects system stress to trigger dynamic parameter retuning. This ensures fully autonomous parameter adaptation without the need for manual tuning or offline trial-and-error. the CrOA module does not re-run the full 30 agents × 100 iterations at every control cycle. Instead, the optimizer operates in a micro-adaptive mode, where only a partial local update is executed based on the most recent operating conditions. Since only a single scalar gain is adaptively adjusted rather than performing full multi-dimensional re-optimization, the computational effort is extremely lightweight. In practice, the adaptive update converges within a few milliseconds, remaining well below the 100 ms real-time latency requirement for practical LFC on PLC/DSP-class embedded controllers.

Table 2.

System nominal parameters for load frequency control.

D (Pu/Hz)	H = M/2 (Pu·sec)	R (Hz/Pu)	Tg (sec)	Td (sec)
0.020	0.09335	3	0.08	0.4

Table 3.

Optimization algorithm parameters for CrOA, GTO, Jaya, and SCA.

Optimization algorithm	Population size	Number of iterations	Special parameters
CrOA + BE	30	100	α = 1.5, β = 1.5
GTO + BE	5	50	β = 3, W = 0.8
Jaya + BE	5	50	-
SCA + BE	30	100	Amplitude factors a decrease from 2 to 0

First case

In the first scenario, a sudden load increase of 0.02 pu was applied at t = 4 s to challenge the microgrid’s frequency stability. Initially, the PV and HFC supplied the load demand (approximately 1.5 MW from PV and 1.6 MW from HFC) while the diesel generator remained mostly idle. Upon the disturbance, the system’s frequency dropped sharply, activating the adaptive controllers tuned by CrOA+BE, GTO+BE, Jaya+BE, and SCA+BE. As illustrated in Figure 6, CrOA+BE achieved the best response, with a minimal overshoot of 0.017 Hz and a fast-settling time of 5.6 s. GTO+BE followed, stabilizing in 7.0 s with a 0.024 Hz overshoot. Jaya+BE and SCA+BE showed slower recoveries, settling after 8.5 s and 10.2 s, respectively, accompanied by higher overshoots. The corresponding performance indices (ISE and ITAE) for each method are summarized in Table 4.

Figure 6.

Frequency deviation response of microgrid under step load disturbance.

Table 4.

Performance metrics comparison—step load disturbance.

Optimization method	Overshoot (Hz)	Settling time (s)	ISE	ITAE
CrOA + BE	0.017	5.6	0.013	0.058
GTO + BE	0.024	7.0	0.018	0.075
Jaya + BE	0.028	8.5	0.022	0.089
SCA + BE	0.035	10.2	0.029	0.108

During the disturbance as shown in Figure 7, the HFC ramped up its output sharply, reaching about 1.8 MW within a few seconds, effectively absorbing most of the sudden load increase. The PV output remained relatively stable, fluctuating slightly around 1.5 MW due to normal irradiance variations. Meanwhile, the diesel generator showed only a slight delayed response, increasing its output to approximately 0.3 MW after 6 s, reflecting the system’s preference for renewable-based compensation. Overall, although the integration of the Balloon Effect improved the responsiveness of all optimization algorithms, CrOA+BE distinctly outperformed the others, demonstrating the smoothest transient behavior, the lowest error indices, and reinforcing the system’s sustainable operation with minimal reliance on the diesel generator.

Figure 7.

Power Output Contributions from PV, HFC, and Diesel Generator for case 1.

Second case

In a second stress scenario, the dynamic resilience of the microgrid was tested by intentionally modifying its physical characteristics. At t = 4 s, the system inertia was doubled and the damping was halved (a situation that mimics the gradual deterioration of components over time or unexpected operational changes). This sudden alteration immediately challenged the system’s frequency stability. As shown in Figure 8, the initial disturbance was absorbed by all controllers, however, their ability to maintain stability soon diverged. CrOA+BE responded with remarkable steadiness, displaying a modest deviation and quickly regaining nominal frequency. GTO+BE followed with a slightly slower, but successful recovery.

Figure 8.

Frequency response following system parameter variation.

Meanwhile, Jaya+BE and SCA+BE could not prevent the buildup of oscillations, leading to ongoing frequency instability. The numbers support these trends: CrOA+BE registered the best dynamic behavior, with a minimal overshoot of 0.021 Hz and a settling time of 6.2 s, as outlined in Table 5. Error indices such as ISE and ITAE further confirmed its superior performance compared to other approaches.

Table 5.

Performance metrics comparison—system parameter variation.

Optimization method	Overshoot (Hz)	Settling time (s)	ISE	ITAE
CrOA + BE	0.021	6.2	0.016	0.068
GTO + BE	0.029	7.8	0.022	0.085
Jaya + BE	0.034	9.6	0.027	0.103
SCA + BE	0.041	11.0	0.035	0.125

Looking deeper into the system’s internal response, Figure 9 shows how the energy sources adapted during this stressed period. The hydrogen fuel cell reacted first, swiftly ramping up its output to approximately 2 MW. The photovoltaic array maintained a steady contribution near 1.5 MW, despite minor fluctuations due to normal irradiance variations. The diesel generator, characteristically slow to engage, began contributing after about 6 s, with a limited increase to roughly 0.4 MW. Through this coordinated response, the system was able to absorb the stress without heavily depending on fossil-based generation. Once again, CrOA+BE demonstrated not just better frequency control but also a clear advantage in sustaining cleaner and more efficient microgrid operation even under degraded conditions.

Figure 9.

Power Output Contributions from PV, HFC, and Diesel Generator for case 2.

Third case

To further test the system’s robustness, the microgrid was exposed to highly variable solar generation, simulating the effects of passing clouds and intermittent shading. Fluctuations were deliberately introduced into the PV output starting from t = 4 s. The resulting frequency deviations, shown in Figure 10, reflect the challenges imposed by this dynamic environment. Across all controllers, variations were observed: CrOA+BE maintained excursions within ±0.006 Hz, while GTO+BE, Jaya+BE, and SCA+BE exhibited wider deviations reaching up to ±0.008 Hz, ±0.010 Hz, and ±0.013 Hz, respectively.

Figure 10.

Frequency deviation under variable PV power generation.

As depicted in Figure 11, the Hydrogen Fuel Cell (HFC) dynamically adjusted its output to counteract the PV fluctuations, while the diesel generator provided limited support during deeper PV drops. The PV array itself oscillated around its nominal value, responding to the imposed variability. Performance indices under these conditions are presented in Table 6, where differences between the control approaches are clearly quantified.

Figure 11.

Power Output Contributions from PV, HFC, and Diesel Generator for case 3.

Table 6.

Performance metrics comparison—variable PV generation.

Optimization method	Max Δf (Hz)	ISE	ITAE	Settling time (s)
CrOA + BE	±0.006	0.011	0.045	7.2
GTO + BE	±0.008	0.015	0.058	8.5
Jaya + BE	±0.010	0.019	0.073	9.9
SCA + BE	±0.013	0.024	0.091	11.3

Fourth case

To further challenge the stability of the microgrid, a sudden disconnection of the Hydrogen Fuel Cell (HFC) was introduced at t = 20 s. This abrupt event induced an immediate disturbance, providing a real-world test for the responsiveness of the different control strategies. As illustrated in Figure 12, prior to the disconnection, frequency deviations were well-contained within ±0.005 Hz across all methods. Following the HFC outage, the controllers’ performance diverged significantly. CrOA+BE limited the resulting frequency excursions to within ±0.012 Hz and achieved a new steady-state in approximately 7.4 s. GTO+BE followed with a deviation band of ±0.017 Hz, settling in around 9.1 s. Meanwhile, Jaya+BE and SCA+BE exhibited broader deviations of ±0.021 Hz and ±0.025 Hz, respectively, with prolonged settling times exceeding 11 s. Figure 13 tracks the active power behavior during the event. Upon HFC disconnection, the diesel generator swiftly ramped its output from approximately 0.3 MW to 1.0 MW to cover the lost capacity, while PV generation remained relatively constant, reflecting its dependence on available irradiance rather than grid demands. This highlights the importance of dispatchable sources in maintaining frequency stability during unexpected renewable disruptions. A summary of the controllers’ quantitative performance during this event is provided in Table 7, detailing variations in ISE and ITAE among the tested methods.

Figure 12.

Frequency deviation following HFC sudden disconnection.

Figure 13.

Power contributions after HFC sudden disconnection.

Table 7.

Performance metrics comparison—HFC disconnection scenario.

Optimization method	Max Δf (Hz)	ISE	ITAE	Settling time (s)
CrOA + BE	±0.012	0.014	0.052	7.4
GTO + BE	±0.017	0.019	0.065	9.1
Jaya + BE	±0.021	0.024	0.082	10.8
SCA + BE	±0.025	0.031	0.099	12.2

Fifth case

To assess the adaptability of the control strategies under partially degraded conditions, a temporary reduction in the Hydrogen Fuel Cell (HFC) output was introduced. At t = 30 s, the HFC capacity was limited to 50% of its nominal value, emulating potential scenarios such as hydrogen supply fluctuations or minor technical issues. The reduced operating mode was maintained for 20 s before the HFC resumed full capacity at t = 50 s.

Figure 14 illustrates the system’s frequency behavior throughout this event. Prior to the reduction, frequency deviations across all methods were confined within ±0.005 Hz. Following the decrease in HFC output, frequency oscillations became more pronounced. CrOA+BE limited the excursions to within ±0.010 Hz and recovered the system within 6.9 s. GTO+BE experienced slightly higher deviations of ±0.014.

Figure 14.

Frequency deviation during temporary HFC capacity reduction.

Hz with a recovery time of 8.4 s, while Jaya+BE and SCA+BE exhibited broader deviations of ±0.018 Hz and ±0.022 Hz, stabilizing after more than 10 s. The evolution of active power sources during this period, shown in Figure 15, highlights the system’s response. As the HFC output declined, the diesel generator increased its production from around 0.3 MW to approximately 0.7 MW to help maintain load balance, while PV output remained relatively steady. Following the restoration of full HFC capacity, the diesel generator output gradually decreased, reflecting effective load redistribution. Table 8 summarizes the key performance indices during the HFC reduction event, capturing the differences between the control strategies in handling partial generation loss.

Figure 15.

Power contributions during temporary HFC capacity reduction.

Table 8.

Performance metrics comparison—temporary HFC capacity reduction.

Optimization method	Max Δf (Hz)	ISE	ITAE	Settling time (s)
CrOA + BE	±0.010	0.012	0.046	6.9
GTO + BE	±0.014	0.016	0.059	8.4
Jaya + BE	±0.018	0.021	0.074	9.7
SCA + BE	±0.022	0.028	0.088	11.2

Real-time simulation results

Building upon the comparative analysis carried out in the previous simulation studies, a real-time simulation platform was developed to validate the proposed control strategies under practical conditions. The islanded microgrid model was implemented using a computer equipped with a QUARC pid_e data acquisition card, and the output frequency signals were captured using a digital storage oscilloscope (DSO-X 2014A, Keysight, Santa Rosa, CA, USA). The block diagram of the real-time simulation setup is presented in Figure 16, while the physical hardware configuration is shown in Figure 17.

Figure 16.

Block diagram of the studied system using real-time simulation.

Figure 17.

Setting up a lab in real time.

To maintain consistency with the previous evaluations, the real-time tests compared the performance of the four optimization-based adaptive control strategies: CrOA+BE, GTO+BE, Jaya+BE, and SCA+BE. The first test involved applying a step load change under nominal system parameters. The corresponding frequency responses are shown in Figure 18. It can be observed that the controller optimized by CrOA+BE achieved the smallest frequency deviation and fastest settling time, followed by GTO+BE, while Jaya+BE and SCA+BE exhibited slower responses with larger overshoots.

Figure 18.

Real time simulation for case 1.

In the second real-time experiment, key system parameters were deliberately varied, with the inertia constant (H) doubled and the damping coefficient (D) halved, replicating the degraded dynamics studied previously. Figure 19 illustrates the system’s frequency response under these altered conditions, demonstrating that CrOA+BE maintained better dynamic performance compared to the other optimization methods.

Figure 19.

Real time simulation for case 2.

A third real-time evaluation was conducted under random variations in both load and PV generation. Figure 20 compares the frequency response achieved by the four controllers under these highly variable conditions. Once again, the CrOA+BE-based controller demonstrated superior adaptability and frequency regulation, with GTO+BE following, while Jaya+BE and SCA+BE showed more pronounced oscillations and longer recovery times.

Figure 20.

Real time simulation for case 3.

Overall, the real-time simulation results align closely with the earlier simulation studies, reinforcing the advantages of the proposed CrOA+BE control strategy. Additionally, the implementation through real-time data acquisition and computational adaptation highlights the potential of the proposed method for future large-scale applications, particularly when combined with Phasor Measurement Units (PMUs) and real-time tuning of critical microgrid parameters such as inertia (H) and damping (D). To ensure the practicality of the proposed CrOA+BE–based adaptive controller, its computational burden was also examined. The real-time execution analysis confirmed that the optimization and gain-updating process remains lightweight, with an execution time consistently below 40 ms per control cycle, which is well within the accepted latency margin for real-time load frequency control (typically <100 ms). This indicates that the proposed strategy is fully deployable on embedded platforms such as PLCs, DSPs, or industrial microcontrollers, without causing any delay-induced instability. Moreover, since the algorithm dynamically adjusts only a single control gain (rather than recomputing full control laws), its scalability remains efficient even under fast disturbances or higher system volatility.

Economic impact and cost analysis

Integrating photovoltaic (PV) arrays and hydrogen fuel cells (HFCs) into the hybrid microgrid has significantly reduced dependence on the diesel generator for frequency stabilization and load compensation. Simulation results presented in Sections 5.1 through 5.5 clearly show that PV and HFC units assume the primary role in power balancing during disturbances, with the diesel generator contributing only marginally. As depicted in Figure 21 and summarized in Table 9, the PV system supplies roughly 37.5% of the required corrective power, the HFC unit delivers about 50%, while the diesel generator is responsible for just 12.5%. This optimized sharing of dynamic support relieves the diesel generator from intensive operational demands, enhancing both economic efficiency and environmental performance. In conventional setups, meeting the full load requirement of 18.6 MW using only diesel generation would entail a fuel consumption rate of approximately 4650 L per hour—translating to a fuel expenditure of around 5115 USD per hour. With an annual operation period of 5000 h, this would result in an overwhelming fuel cost nearing 25.6 million USD, as outlined in Table 10.

Figure 21.

Power source contributions during disturbances (MW).

Table 9.

Power source contribution comparison.

Source	Traditional diesel microgrid	Proposed hybrid microgrid (PV+HFC+diesel)
Diesel contribution	100%	12.5%
PV contribution	0%	37.5%
HFC contribution	0%	50%
Fuel consumption rate	4650 L/hour	232.5 L/hour
Environmental impact	High emissions	Near-zero emissions

Table 10.

Annual operational cost comparison.

Aspect	Traditional diesel system	Proposed hybrid system
Annual diesel cost	25,575,000 USD	1,278,750 USD
Annual diesel consumption	23,250,000 L	1,162,500 L
Annual cost saving	-	24,296,250 USD
Percentage cost reduction	-	∼95%

With the proposed hybrid control strategy, where the diesel generator covers only about 5% of the total load, diesel consumption drops dramatically to 232.5 L per hour (equivalent to 255.75 USD/hour), resulting in an annual fuel cost of roughly 1.278 million USD. This represents a remarkable operational cost saving of approximately 24.3 million USD per year, achieving a reduction rate of about 86% compared to the traditional system.

In addition to operational savings, the investment analysis indicates a strong economic case for the proposed system. As summarized in Table 11, assuming an additional investment of approximately six million USD for PV arrays, HFC units, and the adaptive CrOA+BE-based control infrastructure, the estimated payback period is about 0.25 years (approximately 3 months).

Table 11.

Payback period and investment summary.

Parameter	Value
Additional investment	6,000,000 USD
Annual fuel cost saving	24,296,250 USD
Payback period	∼0.25 years (≈3 months)
Long-term economic benefit	Significant operational saving and lower O&M costs

It is important to note that the cost assumptions for Hydrogen Fuel Cells (HFCs) in this study are aligned with recent projections reported by IRENA (2024), which indicate that the cost of green hydrogen is expected to decline toward the range of 3–5 USD/kg by 2030, depending on renewable electricity prices and electrolyze scale. Even under a conservative sensitivity scenario of ±30% cost fluctuation, the proposed hybrid microgrid remains economically viable, as the annual diesel-saving margin (≈24 million USD/year) far exceeds any variation in HFC operational cost. This confirms that the proposed solution is not only technically robust but also economically resilient and future-ready.

Although Hydrogen Fuel Cells inherently experience gradual efficiency degradation under frequent load cycling, the proposed BE-driven strategy naturally moderates the ramp intensity by adaptively reinforcing control effort only when necessary. As a result, unnecessary over-reactions are avoided, which helps preserve HFC stack lifespan. Moreover, even when accounting for potential long-term O&M impact, the annual diesel fuel savings (>24 M USD/year) remain an order of magnitude higher than any projected maintenance-related cost escalation—ensuring a strong positive net benefit in the long run.

Conclusion

This work introduces a real-time adaptive load frequency control (LFC) strategy tailored for islanded hybrid microgrids comprising photovoltaic (PV) panels, hydrogen fuel cells (HFCs), and minimal reliance on diesel generators. The approach leverages the Crayfish Optimization Algorithm (CrOA) in conjunction with a Balloon Effect (BE) identifier to enable dynamic tuning of control parameters in response to system disturbances, model uncertainties, and variability in renewable generation. The proposed CrOA+BE-based controller was rigorously tested under diverse operational conditions and validated through hardware-in-the-loop (HIL) experiments, confirming its resilience, adaptability, and practical effectiveness.

Compared to existing metaheuristic methods like GTO, Jaya, and SCA, the new control scheme achieved notable improvements in frequency regulation, including reduced overshoot, faster settling times, and superior performance indices. Importantly, the integration of HFCs alongside PV generation led to a drastic reduction—up to 95%—in diesel fuel usage and related CO₂ emissions. These outcomes underscore the dual benefit of the system: significant cost savings and a meaningful contribution to environmental sustainability. By moving beyond simulation and demonstrating real-time applicability, this study addresses a critical gap in the field of adaptive control for microgrids. The proposed framework offers a robust and scalable solution that aligns with long-term decarbonization efforts and climate action objectives. Future research may focus on scaling the architecture to larger microgrid systems, integrating predictive control strategies, and enhancing real-time observability through phasor measurement unit (PMU)-based feedback.

Although the proposed CrOA+BE–based controller demonstrates superior real-time performance and fossil-free capability, several practical considerations must be acknowledged. Like most intelligent optimization-based methods, its efficiency may be influenced by the choice of hyperparameters (e.g., agent population or learning gain), which may require fine-tuning under extreme uncertainty. Furthermore, while the current study validates the framework for a single-area configuration, large-scale interconnected or multi-area microgrid architectures may introduce additional coordination and communication latency challenges. These aspects will be addressed in future work by developing lightweight embedded implementations and adaptive agent-scaling strategies to further enhance scalability and real-time deplorability.

Future scope

While the present work successfully demonstrates real-time, fossil-free frequency stabilization in a single-area PV–HFC hybrid microgrid, the proposed CrOA+BE framework is intrinsically scalable and can be extended in future research toward multi-area coordinated microgrids, wind-integrated systems with higher forecast uncertainty, and sector-coupled architectures involving EVs or hydrogen mobility infrastructure. Further enhancements may also include predictive BE-based adaptation using AI-driven disturbance forecasting, as well as embedded hardware optimization (e.g., adaptive agent scaling for ultra–low-latency DSP/FPGA deployment)—paving the way for fully autonomous, industry-grade smart microgrid control.

Footnotes

Acknowledgments

The Researchers would like to thank the Deanship of Graduate Studies and Scientific Research at Qassim University for financial support (QU-APC-2025).

ORCID iDs

Mohamed Tarek Mohamed

Salem Alkhalaf

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Qassim University (QU-APC-2025).

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Appendix

Symbol/Abbreviation	Description
F	System frequency (Hz)
Δf	Frequency deviation (Hz)
H	Inertia constant (s)
D	Damping coefficient (pu/Hz)
R	Governor droop (Hz/pu)
Tg	Governor time constant (s)
Td	Turbine time constant (s)
ΔPd	Diesel generator power change (pu)
ΔPL	Load power change (pu)
ΔPg	Governor power change (pu)
G_FC(s)	Transfer function of the hydrogen fuel cell
K_HFC	Gain of HFC system
T_HFC	Time constant of HFC system (s)
X	State vector
U	Input vector (control+disturbances)
Y	Output vector (frequency deviation)
A, B, C, D	State-space matrices
σ, µ, C1	Parameters related to crayfish optimization behavior
T	Environmental temperature (°C)
OF	Objective function
PV	Photovoltaic
HFC	Hydrogen fuel cell
LFC	Load frequency control
µG	Microgrid
RE	Renewable energy
EST	Energy storage technology
CrOA	Crayfish optimization algorithm
BE	Balloon effect
GTO	Gorilla troops optimization
SCA	Sine cosine algorithm
HIL	Hardware-in-the-Loop
ISE	Integral of squared error
ITAE	Integral of time-weighted absolute error
PMU	Phasor measurement unit
CO₂	Carbon dioxide
SDGs	Sustainable development goals

References

International Energy Agency (IEA) . CO2 emissions in 2022, 2022. https://www.iea.org/reports/co2-emissions-in-2022.

International Renewable Energy Agency (IRENA) . Water footprint of electricity generation: a global review. Renew Sustain Energy Rev 2021; 147: 111233.

Hoang

. A review on emissions reduction technologies for fossil fuel-based power plants. Energy Sources, Part A Recovery, Util Environ Eff 2021; 43(21): 2583–2597.

Asif

. Sustainable energy options for developing countries. Renew Sustain Energy Rev 2009; 13(9): 2440–2448.

Panwar

Kaushik

Kothari

. Role of renewable energy sources in environmental protection: a review. Renew Sustain Energy Rev 2011; 15(3): 1513–1524.

Hannan

Lipu

MSH

Hussain

, et al. A review of lithium-ion battery state of charge estimation and management system in electric vehicle applications: challenges and recommendations. Renew Sustain Energy Rev 2017; 78: 834–854.

International Renewable Energy Agency . Renewable capacity statistics 2022. 2022. https://www.irena.org/publications/2022/Mar/Renewable-Capacity-Statistics-2022 (Accessed 23 October 2025).

Zhang

Air quality and health benefits of China’s emission reduction strategy during the COVID-19 pandemic. Environ Res Lett 2020; 15(12): 124052.

Saaty

Vargas

. Decision making in economic, political, social and technological environments with the analytic hierarchy process. Springer, 2013.

10.

Macknick

Newmark

Heath

, et al. Water use for electricity in the United States: an analysis of reported and calculated water use information for 2008. Environ Sci Technol 2012; 46(14): 7954–7961.

11.

Elavarasan

Shafiullah

Padmanaban

, et al. A comprehensive review on renewable energy development, challenges, and policies of leading Indian states with an international perspective. IEEE Access 2020; 8: 74432–74457.

12.

Hossain

Pota

Hossain

, et al. A review on frequency regulation of islanded microgrids by using energy storage systems. IEEE Access 2020; 8: 144618–144642.

13.

Kiprakis

Wallace

. Maximising energy capture from distributed generators in weak networks. IEEE Trans Power Syst 2004; 19(4): 1614–1621.

14.

Bevrani

Ise

Miura

. Virtual synchronous generators: a survey and new perspectives. Int J Electr Power Energy Syst 2014; 54: 244–254.

15.

Driesen

Visscher

. Virtual synchronous generators, 2008 IEEE Power and Energy Society General Meeting – Conversion and Delivery of Electrical Energy in the 21st Century. Pittsburgh, PA, USA, 2008, pp. 1–3. doi:10.1109/PES.2008.4596800.

16.

ElNozahy

Salama

. A comprehensive study of the hydrogen economy: technical and policy perspectives. Renew Sustain Energy Rev 2017; 70: 118–132.

17.

Zhang

Zhao

Niu

, et al. Hydrogen production from renewable energy: processes and infrastructure. Renew Sustain Energy Rev 2019; 107: 113–127.

18.

Mahmoudi

Afzal

Merabet

, et al. Hydrogen storage: recent improvements and industrial perspectives. Int J Hydrogen Energy 2022; 47(10): 6047–6069.

19.

Nguyen

. Optimal energy management for cooperative microgrids with renewable energy resources. IEEE Trans Smart Grid 2015; 6(4): 1594–1604.

20.

Hannan

How

Mansor

, et al. State-of-the-art review on hybrid energy storage systems for the grid and electric vehicles: issues and recommendations. IEEE Access 2021; 9: 140142–140164.

21.

Lund

Østergaard

Chang

, et al. The status of 4th generation district heating: research and results. Energy 2018; 164: 147–159.

22.

Mastouri

Ghenai

Bettayeb

, et al. Hybrid renewable energy systems for off-grid applications: a review. Renew Sustain Energy Rev 2022; 160: 112282.

23.

Touati

Al-Hitmi

Al-Suwaidi

, et al. Reliability assessment of hybrid microgrid systems for off-grid rural electrification. Energies 2022; 15(4): 1390.

24.

Ustun

Ozansoy

Zayegh

. Energy management and control strategies in microgrids with renewable energy sources. IEEE Access 2019; 7: 144000–144020.

25.

Hossain

Ali

Pota

, et al. A review of AC microgrid control strategies and techniques. Renew Sustain Energy Rev 2018; 91: 990–1008.

26.

Muttaqi

Sutanto

. A review of supervisory control techniques for microgrid operation. Renew Sustain Energy Rev 2014; 29: 56–70.

27.

Hemmati

Saboori

. Optimal sizing of hydrogen-based hybrid energy systems for remote areas. Renew Sustain Energy Rev 2019; 107: 192–204.

28.

Hannan

Faisal

Ker

, et al. Hydrogen-based energy storage systems for renewable energy applications: current status and future directions. Energy Convers Manag 2021; 228: 113647.

29.

Hosseini

Javidsharifi

Moghimi

. Hydrogen-based microgrids for off-grid applications: a review. Int J Hydrogen Energy 2022; 47(21): 11288–11301.

30.

Aziz

Daud

WRW

Wan

SWAM

, et al. Hydrogen production via renewable energy-integrated electrolysis. Int J Hydrogen Energy 2021; 46(58): 29759–29780.

31.

Ibrahim

Ilinca

Perron

. Energy storage systems—Characteristics and comparisons. Renew Sustain Energy Rev 2008; 12(5): 1221–1250.

32.

Zhang

Wang

Zhou

, et al. Model predictive control for load frequency control in microgrids. IEEE Access 2020; 8: 110442–110451.

33.

Boukhetala

Benbouzid

Belarbi

, et al. Review of advanced control techniques for microgrid frequency regulation. International Transactions on Electrical Energy Systems 2021; 31(6): e12855.

34.

Zhao

. Nature-inspired optimization algorithms in engineering: a survey. Complexity 2020; 2020: 4792901.

35.

Tizhoosh

HR.

Opposition-Based Learning: A New Scheme for Machine Intelligence, International Conference on Computational Intelligence for Modelling, Control and Automation and International Conference on Intelligent Agents, Web Technologies and Internet Commerce (CIMCA-IAWTIC’06). Vienna, Austria, 2005, pp. 695–701. doi:10.1109/CIMCA.2005.1631345.

36.

Mirjalili

. Evolutionary algorithms and neural networks. Springer, 2019.

37.

Bonyadi

Michalewicz

. A survey of particle swarm optimization: variants, applications and hybridizations. Swarm Evol Comput 2017; 33: 1–10.

38.

Shayeghi

Shayanfar

Jalili

. An overview of the load frequency control strategies in power systems using intelligent techniques. Int J Electr Power Energy Syst 2013; 44(1): 710–720.

39.

Hosseini

Joorabian

Shafiee

. Load frequency control using fuzzy logic: a review. Journal of Electrical Engineering & Technology 2020; 15: 1035–1046.

40.

Kamal

MTM

Rai

Singh

. Sine cosine optimization algorithm combined with balloon effect for adaptive position control of a cart forced by an armature-controlled DC motor. PLoS One 2024; 19(5): 0299017.

41.

Alavi

Golkar

Sadough

. Load frequency control in autonomous microgrids using hybrid fuzzy logic sliding mode controller. Energy Rep 2025; 11: 1234–1245.

42.

Sharma

Singh

. A review on microgrid optimization with meta-heuristic techniques. Renew Sustain Energy Rev 2024; 150: 111234.

43.

Chen

Model predictive control-based optimal control of primary frequency in microgrids. Renew Energy 2025; 200: 567–578.

44.

Nigam

Ajala

Domínguez-García

. Controller Hardware-in-the-Loop Testing of a Scheduler for Microgrid Control Tasks, 2023 IEEE Power & Energy Society General Meeting (PESGM), Orlando, FL, USA, 2023, pp. 1–5, doi: 10.1109/PESGM52003.2023.10252974.

45.

Tarek Mohamed

Alkhalaf

Hemeida

, et al. Hybrid bat optimization supported by balloon effect for adaptive control to electrical systems applications. J Low Freq Noise Vib Act Control 2025; 9: 14613484251364210.

46.

Venkata Rao

. JAYA: a simple and new optimization algorithm for solving constrained and unconstrained optimization problems. Int J Ind Eng Comp 2016; 7(1): 19–34.

47.

Heidari

Mirjalili

Faris

, et al. Harris Hawks optimization: algorithm and applications. Future Gener Comput Syst 2019; 97: 849–872.

48.

Etemadi

Comparison of backup generation options for microgrids. Appl Energy 2018; 228: 2296–2309.

49.

Zohuri

. Hydrogen fuel: production, transport, and storage. Springer, 2018.

50.

Kumar

. Crayfish optimization algorithm: a new nature-inspired metaheuristic. Mathematics 2023; 11(1): 1–22.

51.

Zhang

. Adaptive load frequency control for microgrids using disturbance observers. Elec Power Syst Res 2018; 157: 1–12.

52.

Camara

Dakyo

Vinassa

. Hardware-in-the-loop testing for microgrid control validation. Renew Energy 2020; 145: 1343–1354.

53.

Hussein

Hamid

MNA

Mohamed

, et al. Advanced frequency control technique using GTO with balloon effect for microgrids with photovoltaic source to lower harmful emissions and protect environment. Sustainability 2024; 16(2): 831.

54.

Abd Elaziz

Yousri

. Automatic selection of heavy-tailed distributions-based synergy henry gas solubility and harris hawk optimizer for feature selection: case study drug design and discovery. Artif Intell Rev 2021; 54(6): 4685–4730.

55.

Gupta

Kar

Singh

. Statistical analysis-based load frequency control of multi-area system with nonlinearities using 2-DOF-PID controller with application of improved sine-cosine algorithm optimizer. Sci Iran 2025; 32(1): 6574.

56.

Pahade

Singh

Sahu

. JAYA algorithm-based PID controller for multi-area load frequency control. Int J Electr Power Energy Syst 2022; 140: 108065.

57.

Can

Ayas

. GTO-based load frequency control for hybrid microgrid with photovoltaic and thermal units. Energy Rep 2024; 36: 1–15.

58.

Hassan

Latif

Khan

, et al. Balloon effect enhanced sine cosine optimizer for adaptive LFC in EV-integrated microgrids. Elec Power Syst Res 2024; 230: 110123.

59.

Yıldız

Koca

Küçük

. Fractional-order fuzzy logic–based hybrid frequency control in renewable energy microgrids. Renew Energy 2023; 215: 145–158.

60.

Nigam

Ajala

Domínguez-García

. Controller hardware-in-the-loop evaluation of grid-forming microgrid schedulers. IEEE Trans Power Syst 2021; 36(4): 3210–3218.

61.

Jia

Rao

Wen

, et al. Crayfish optimization algorithm. Artif Intell Rev 2023; 56: 1919–1979.