---
title: Improved Golden Eagle Optimization (IGEO)
url: https://www.emergentmind.com/topics/improved-golden-eagle-optimization-igeo
type: topic
---

# Improved Golden Eagle Optimization (IGEO)

Improved Golden Eagle Optimization (IGEO) is a discrete, genetic-enhanced version of the continuous Golden Eagle Optimization (GEO) algorithm, tailored to solve the task scheduling problem in fog computing under deadline and energy constraints. Within the broader Reinforcement Improved Golden Eagle Optimization framework, or RIGEO, fog nodes are classified by network traffic into low-traffic and high-traffic categories, tasks are classified by deadline into low-deadline and high-deadline categories, and IGEO serves as the metaheuristic engine for scheduling low-deadline tasks on low-traffic fog nodes, while reinforcement learning is used for high-deadline tasks on high-traffic nodes [2509.07378].

## 1. Position within deadline-aware fog scheduling

IGEO is defined in the setting of deadline-aware task scheduling for Internet of Things workloads executed over fog infrastructure. The scheduling objective is to decrease energy usage and improve response times for application requests while taking task deadlines into account. The fog layer is organized around heterogeneous fog nodes, and the scheduling problem is cast as the assignment of tasks to nodes under deadline and energy considerations.

The defining architectural decision is the split of responsibilities inside RIGEO. Fog nodes are dynamically classified according to average network traffic, and tasks are routed according to deadline urgency. If a task deadline is below a deadline threshold, the scheduler selects from low-traffic fog nodes and invokes `IGEO_PROCESS(T, fc)`; otherwise, it selects from high-traffic fog nodes and invokes `RL_PROCESS(T, fc)` [2509.07378].

This division gives IGEO a sharply delimited role. It is not presented as a general-purpose scheduler for all fog workloads. Rather, it is specialized for low-deadline tasks that require fast, good-quality schedules with limited queueing delay. A plausible implication is that the algorithmic design of IGEO is shaped less by generic global optimization concerns than by the operational need to make high-quality discrete scheduling decisions quickly in the low-traffic portion of the system.

## 2. Optimization model and performance criteria

The underlying decision problem is to map a task set
\[
T = \{t_1, t_2, \ldots, t_n\}
\]
to a fog-node set
\[
F = \{f_1, f_2, \ldots, f_m\}
\]
through a mapping
\[
M : T \rightarrow F.
\]

Each task \(t_i\) has a deadline \(d_i\) in milliseconds. The response time of task \(t_i\) is modeled as
\[
R_i = P_i + T_i + E_i + Q_i,
\]
where \(P_i\) is propagation delay, \(T_i\) is transmission delay, \(E_i\) is execution or processing time on the CPU, and \(Q_i\) is waiting time in the queue on the assigned fog node [2509.07378].

Deadline satisfaction is represented through per-task and aggregate violation terms:
\[
DV_i = \max(0, R_i - d_i),
\]
\[
DV_{\text{total}} = \sum_{i=1}^{n} DV_i.
\]

Energy consumption is defined at the fog-node level as
\[
E_j = \alpha \cdot E_{\text{active},j} + \beta \cdot E_{\text{idle},j},
\]
with total energy
\[
E_{\text{total}} = \sum_{j=1}^{m} E_j.
\]

The conceptual optimization goals are to minimize total energy consumption, minimize response time, and minimize total deadline violation. The exact combined scalar objective function is not explicitly written; the paper states only that these terms are “systematically integrated into the fitness function.” An inferred structure is a weighted scalarization over \(E_{\text{total}}\), \(\sum_i R_i\), and \(DV_{\text{total}}\), but the weights and exact formula are not specified. Likewise, constraints such as one-node-per-task assignment and finite CPU and bandwidth limits are implied rather than written explicitly.

This formulation places IGEO within a standard multi-criteria fog-scheduling setting, but with incomplete formal specification. The omission is important because it affects reproducibility and makes the precise tradeoff among energy, latency, and deadline satisfaction dependent on implementation details not fully exposed in the text.

## 3. From continuous GEO to discrete IGEO

The original GEO is a continuous swarm-intelligence optimizer based on an “attack + cruise” movement model. For eagle \(i\), the attack vector is
\[
\vec{A}_i = \vec{X}_f^* - \vec{X}_i,
\]
and the step vector is
\[
\Delta x_i = \vec{r}_1\, p_a \frac{\vec{A}_i}{\lVert \vec{A}_i \rVert} + \vec{r}_2\, p_c \frac{\vec{C}_i}{\lVert \vec{C}_i \rVert},
\]
where \(\vec{r}_1\) and \(\vec{r}_2\) are random vectors with elements in \([0,1]\), \(p_a\) is the attack coefficient, \(p_c\) is the cruise coefficient, and \(\vec{C}_i\) is the cruise vector. The continuous GEO position update is
\[
x^{(t+1)} = x^{(t)} + \Delta x_i^{(t)}.
\]

IGEO modifies this scheme in two principal ways. First, it replaces direct continuous motion with discretization via genetic operators, because task scheduling is a discrete combinatorial problem. Positions of eagles are interpreted as discrete solutions, specifically task-to-node assignment strings, and mutation and crossover from Genetic Algorithms become the main update operators. Second, it preserves GEO’s exploration-exploitation logic by using the sign and relative magnitudes of the step vector as a control signal for choosing which genetic operator to apply [2509.07378].

| GEO regime | IGEO operator | Search role |
|---|---|---|
| Cruise dominant or negative step | Mutation | Exploration |
| Attack dominant or positive step | Crossover | Exploitation |

This design preserves the conceptual structure of GEO while changing its execution model from continuous displacement to discrete combinational search. The intended effect is better diversification when GEO signals exploration and better intensification around good schedules when GEO signals exploitation.

## 4. Operator selection, update rules, and algorithmic workflow

IGEO computes the same GEO-style step information and then converts it into a discrete update rule. The decision logic is described in two equivalent ways. If \(|r_1 p_a| < |r_2 p_c|\), cruise dominates and mutation is used; if \(|r_1 p_a| > |r_2 p_c|\), attack dominates and crossover is used. The paper also phrases the rule in terms of sign: a negative step corresponds to exploration and triggers mutation, whereas a positive step corresponds to exploitation and triggers crossover.

When the step vector is negative, the update rule is
\[
x_{t+1} =
\begin{cases}
\text{Mutation}(x_{\text{best}}) & \text{if } r \geq 0.5,\\
\text{Mutation}(x_t) & \text{if } r < 0.5.
\end{cases}
\]

When the step vector is positive, the update rule is
\[
x_{t+1} =
\begin{cases}
\text{single-point crossover}(x_{\text{best}}, x_t) & \text{if } r \geq 0.5,\\
\text{two-point crossover}(x_{\text{best}}, x_t) & \text{if } r < 0.5.
\end{cases}
\]

The mutation operator is described conceptually as random alteration of parts of the parent’s genetic material. The crossover operators are standard one-point and two-point crossovers. The paper does not specify mutation rate, how many genes are mutated, or detailed gene-selection rules.

An inferred internal workflow of IGEO consists of initialization of a population of candidate schedules, evaluation through the fitness function, identification of a global best schedule, iterative computation of attack and cruise vectors, operator selection from the GEO-derived step information, evaluation of the new schedule, and updating of personal and global bests. Termination is by a maximum iteration count or detected convergence, although no explicit iteration count, population size, or stopping criterion is provided [2509.07378].

Parameterization is correspondingly partial. Explicit or implied parameters include the attack coefficient \(p_a\), the cruise coefficient \(p_c\), random vectors \(\vec{r}_1, \vec{r}_2 \sim U(0,1)\), the \(0.5\) threshold for choosing whether mutation acts on \(x_{\text{best}}\) or \(x_t\), and the \(0.5\) threshold for choosing single-point versus two-point crossover. Population size, number of iterations, and detailed stopping conditions are not stated.

## 5. Empirical behavior in fog computing experiments

The reported experiments evaluate RIGEO as a whole rather than IGEO in isolation, but IGEO is a major component of the low-deadline scheduling path. The comparison set includes standard GEO, GWO, WCLA+GA, and ETFC. The evaluation uses 200 to 600 IoT tasks, 20 heterogeneous fog nodes, CPU capacities of 2000–6000 MIPS per node, node energy per operation randomly drawn from 80–200 Joules, MATLAB 2016 on Windows 10 with an Intel i7 processor and 16 GB RAM, and 50 runs with parallel processors [2509.07378].

Three metrics are emphasized: total energy consumption, total deadline violation time, and total response time. The results are described qualitatively rather than numerically. RIGEO is reported to use less energy than GEO, GWO, WCLA+GA, and ETFC as the number of tasks grows; to reduce total deadline violation time significantly; and to produce the lowest total response time among the compared algorithms.

The paper attributes these improvements to better task distribution among nodes, including IGEO’s handling of low-deadline tasks on low-traffic nodes. In particular, the assignment of time-critical tasks to low-traffic nodes is intended to reduce queueing time \(Q_i\), while IGEO’s fast metaheuristic search is intended to reduce deadline violations for those tasks.

Several limits of the evaluation are explicit. The experiments do not separate the performance of IGEO alone from that of the full RIGEO system, no numerical percentages or statistical tests are given, and the comparisons are graphical and descriptive rather than inferential. Accordingly, the empirical evidence supports the effectiveness of the hybrid system more directly than it isolates the marginal contribution of IGEO.

## 6. Assumptions, misconceptions, limitations, and related improved GEO variants

The system model assumes a three-layer IoT–Fog–Cloud architecture, independent tasks with known deadlines, heterogeneous fog nodes with CPU capacities of 2000–6000 MIPS, and traffic-based classification of nodes via comparison with a threshold. These assumptions simplify the representation and focus the optimization on mapping tasks to nodes.

Two common misconceptions are directly addressed by the formulation. First, IGEO is not the complete RIGEO scheduler; it is only the component responsible for low-deadline tasks on low-traffic fog nodes. Second, the method is not fully specified at the level of reproducible optimization details: the exact scalar fitness function, its weights, several GEO and GA parameter values, and the exact reinforcement-learning algorithm, state space, action space, and reward formula are not provided.

Limitations noted in the source include incomplete formalization, evaluation only up to 20 fog nodes and 600 tasks, omission of formal complexity analysis for IGEO, and unresolved handling of sudden, unpredictable traffic spikes. The paper also remarks that metaheuristic algorithms introduce overhead and therefore confines IGEO to the low-traffic regime, while RL is described as having higher time complexity and is reserved for high-deadline tasks.

In a broader methodological context, IGEO belongs to a family of improved GEO variants that retain the attack–cruise logic of GEO while incorporating evolutionary operators. A directly related example is Golden Eagle Genetic Optimization (GEGO), which integrates GEO with selection, crossover, and mutation inside the iterative search process rather than as a separate stage. GEGO preserves GEO’s movement equations and uses periodic population-wide genetic interventions to improve diversity and reduce premature convergence, thereby illustrating the same general principle of hybridizing GEO dynamics with genetic search mechanisms [2601.14672].

This broader comparison suggests that IGEO exemplifies a recurring design pattern in improved GEO research: the dynamics of a continuous swarm optimizer are preserved as a controller of exploration and exploitation, but the actual search move is re-expressed in operators suited to the target search space. In IGEO, that search space is discrete fog-task assignment under deadline, response-time, and energy criteria.

Source: https://www.emergentmind.com/topics/improved-golden-eagle-optimization-igeo