💻 computer science

Internal Combustion Engine Optimization: An Improved Physics-Inspired Metaheuristic

This paper introduces Internal Combustion Engine Optimization (ICEO), a novel metaheuristic algorithm inspired by the four-stroke engine cycle and governed by an adaptive Engine Control Unit, which significantly outperforms state-of-the-art methods in solving high-dimensional, nonlinear optimization problems by effectively balancing exploration and exploitation to avoid premature convergence.

Original authors: Chabi A. R. Babatounde, Bastien Poggi, Evelyne Vittori, Thierry Antoine-Santoni, Antoine Aiello

Published 2026-07-01

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Chabi A. R. Babatounde, Bastien Poggi, Evelyne Vittori, Thierry Antoine-Santoni, Antoine Aiello

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to find the absolute best route through a massive, foggy, and constantly shifting maze. You have a team of explorers (a computer program) looking for the exit. This is what engineers call an "optimization problem."

For decades, computer scientists have used "metaheuristics"—smart guessing strategies inspired by nature (like ants finding food or birds flocking)—to solve these puzzles. But these strategies often get stuck. They might find a small, easy valley and think it's the bottom of the whole world, missing the deep, perfect valley hidden behind a mountain. This is called "premature convergence."

This paper introduces a new strategy called ICEO (Internal Combustion Engine Optimization). Instead of copying nature, the authors copied a car engine.

The Big Idea: Your Algorithm is a Car Engine

The authors realized that a four-stroke car engine is actually a perfect machine for searching. It doesn't just do one thing; it cycles through four distinct, coordinated phases to turn fuel into power. ICEO does the same thing to turn "random guesses" into the "best solution."

Here is how the four phases work, using the engine analogy:

Intake (Breathing In):
- The Engine: The piston pulls in fresh air and fuel.
- The Algorithm: The computer pulls in a mix of "fresh air" (brand new, random guesses) and "used air" (solutions it found earlier). This ensures the team keeps exploring new areas of the maze so they don't get stuck in one spot.
Compression (Squeezing):
- The Engine: The piston squeezes the air-fuel mix, making it dense and hot.
- The Algorithm: The computer takes the promising solutions and "squeezes" them closer together. It focuses the search on the most interesting areas it has found so far, getting ready for a big breakthrough.
Combustion (The Explosion):
- The Engine: The spark plug ignites the mix, creating a powerful explosion that pushes the piston.
- The Algorithm: This is the "magic moment." The computer takes the squeezed solutions and applies a sudden, intense burst of change (like a spark). This helps the team jump out of small traps (local optima) and potentially land in a much better spot.
Exhaust (Cleaning Out):
- The Engine: The piston pushes out the burnt, useless gases.
- The Algorithm: The computer throws away the bad, weak solutions and keeps only the best ones to start the next cycle.

The Brain: The "ECU" (Engine Control Unit)

In a real car, a computer chip called the ECU monitors the engine. If the engine is struggling, it adjusts the fuel mix. If it's running hot, it cools it down.

ICEO has its own ECU. This is the smartest part of the paper.

Most algorithms just follow a fixed plan (e.g., "explore for 50% of the time, then exploit for 50%").
ICEO's ECU watches the engine in real-time. It uses a "sliding window" to look at the last few steps.
- If the team is stuck and not improving (stagnation), the ECU says, "Okay, we need more fresh air!" and injects more random guesses to shake things up.
- If the team is making great progress, the ECU says, "Great, let's squeeze harder and focus!" to refine the best solution.

This makes the algorithm adaptive. It doesn't just follow a script; it reacts to the problem as it happens.

Did It Work?

The authors tested ICEO against 10 other famous algorithms (like Particle Swarm Optimization and Genetic Algorithms) using two types of tests:

Math Mazes (Benchmarks): They used 18 standard math problems and 12 very difficult, modern problems (CEC 2022 suite).
- Result: ICEO won the most often. It found the best solutions more consistently than the others, especially on the hardest, most complex mazes.
- Statistical Proof: They used strict math tests (Wilcoxon and Nemenyi tests) to prove that ICEO wasn't just getting lucky. It was statistically better than many of the competitors, though it was very close to a few top-tier rivals like L-SHADE.
Real-World Engineering: They tried to solve three classic engineering problems:
- Speed Reducer: Designing a gearbox to be as light as possible.
- Three-Bar Truss: Designing a bridge structure to be light but strong.
- Pressure Vessel: Designing a tank to hold gas safely.
- Result: ICEO performed just as well as the best existing methods, finding optimal designs with high stability.

The "Ablation" Test (Taking it apart)

To prove that the engine analogy actually mattered, the authors broke the algorithm apart:

They removed the "Compression" phase: The algorithm got much worse.
They removed the "ECU" (the brain): The algorithm got worse, proving that the ability to adapt in real-time is crucial.
Conclusion: The magic isn't just in one part; it's in how the four phases work together under the supervision of the ECU.

The Bottom Line

This paper proposes a new way to solve complex problems by treating the search process like a car engine cycle. By constantly breathing in new ideas, squeezing them, exploding them with energy, and cleaning out the trash—all while a smart "ECU" adjusts the settings on the fly—the algorithm avoids getting stuck and finds better answers faster than many current methods.

What the paper does NOT claim:

It does not claim this will fix actual car engines.
It does not claim this works for medical diagnoses or clinical uses.
It does not claim it is perfect for every problem (the "No Free Lunch" theorem still applies), but it is a very strong tool for complex, high-dimensional engineering and math problems.

Technical Summary: Internal Combustion Engine Optimization (ICEO)

Problem Statement
Optimization in high-dimensional, nonlinear, and multimodal search spaces remains a fundamental challenge in engineering and computer science. While metaheuristics have proven effective for NP-hard problems, many existing algorithms suffer from premature convergence and an inability to maintain an optimal balance between exploration (diversification) and exploitation (intensification). Current adaptive approaches often rely on random mutation or perturbation strategies that do not explicitly incorporate the search process's historical dynamics or current state, leading to stagnation in local optima.

Methodology
The paper proposes Internal Combustion Engine Optimization (ICEO), a physics-inspired metaheuristic structured around the four-stroke operating cycle of an internal combustion engine. Unlike purely metaphorical algorithms, ICEO is designed as a structured, adaptive system where four coordinated phases interact synergistically:

Intake (Exploration): Corresponds to the generation and integration of diverse new solutions. The algorithm draws "air" (candidate solutions) from the atmosphere (population) and "fuel" (random perturbation vectors). The proportion of fresh air versus recycled "burned gases" (previously discarded solutions) is dynamically regulated.
Compression (Early Exploitation): The search space is progressively contracted around the current best solution ( $X_{best}$ ). Air particles are combined with fuel vectors to generate new candidate configurations, focusing the search on promising regions.
Combustion (Intensive Exploitation & Escape): Modeled as a probabilistic process where only a subset of solutions undergoes "ignition." This phase involves intensive refinement of the best solutions. To prevent entrapment in local minima, the algorithm occasionally applies Lévy flight-based perturbations to selected solutions.
Exhaust (Selection & Update): Burned gases (newly generated solutions) are evaluated. The population is sorted, and the best solutions are retained for the next cycle, while inferior candidates are expelled.

Adaptive Control Mechanism (ECU)
A central innovation of ICEO is the Engine Control Unit (ECU), an adaptive mechanism that governs the iterative cycle. The ECU monitors the convergence speed and stagnation using a sliding window ( $\omega$ ) of recent fitness values.

It calculates the average improvement rate ( $\Delta f$ ) and adjusts an acceleration parameter ( $a_x$ ).
Based on $a_x$ , the ECU dynamically determines the air-fuel ratio ( $\rho_{air}$ and $\rho_{fuel}$ ), effectively controlling the balance between injecting new diversity (exploration) and refining existing solutions (exploitation).
This feedback-driven adaptation allows the algorithm to react to the optimization dynamics in real-time rather than relying on fixed schedules.

Key Contributions

Structured Four-Phase Cycle: ICEO explicitly separates search operations into distinct, coordinated stages (intake, compression, combustion, exhaust), differing from algorithms that embed these functions within single update rules.
Adaptive Regulation: The introduction of the ECU mechanism allows for performance-dependent parameter adjustment, specifically tuning the intensity of the "combustion" process and the rate of solution injection based on observed stagnation.
Air-Fuel Mixture Mechanism: The algorithm explicitly decomposes search resources into external diversity injection (fuel) and population-based continuity (air), with the ratio adaptively regulated to balance exploration and exploitation.

Experimental Results
The performance of ICEO was evaluated against ten state-of-the-art metaheuristics (including L-SHADE, CMA-ES, GWO, PSO, and GA) across three experimental stages:

Standard Benchmark Functions (D=100): Tested on 18 functions (unimodal and multimodal). ICEO achieved the lowest global rank score (27 points), securing 15 first-place finishes. It demonstrated superior convergence accuracy on both unimodal (e.g., Dixon-Price) and multimodal (e.g., Rastrigin, Ackley) functions compared to competitors.
CEC 2022 Benchmark Suite (D=20): Evaluated on 12 complex functions (unimodal, multimodal, hybrid, and composition). ICEO achieved the best average rank (2.750) in the Friedman test.
- Statistical Significance: The Nemenyi post-hoc test indicated that ICEO's superiority was statistically significant only against the lowest-performing algorithms (WarSO, EHO, CMA-ES). Differences against top competitors like L-SHADE and SADE were not statistically significant under the Nemenyi test, though the Wilcoxon signed-rank test showed ICEO winning significantly more pairwise comparisons (94 wins vs. 18 losses overall).
Real-World Engineering Problems: ICEO was applied to the Speed Reducer, Three-Bar Truss, and Pressure Vessel design problems. In all cases, it demonstrated competitive performance, fast convergence, and low solution dispersion compared to the top five algorithms from the benchmark tests.
Ablation Study: Removing the compression phase (V1) caused the most significant performance degradation, highlighting its critical role. Variants without the ECU (V3) performed worse than the full ICEO, confirming the value of the adaptive mechanism.

Significance and Limitations
The paper claims that ICEO offers a robust framework for balancing exploration and exploitation through its structured, physics-inspired cycle and real-time adaptive control. The ECU mechanism distinguishes it from standard leader-guided algorithms by adapting search effort based on performance rather than time.

However, the authors acknowledge specific limitations:

Statistical Nuance: While ICEO often ranks first, the Nemenyi test suggests its superiority is not statistically confirmed against all top-tier competitors (e.g., L-SHADE), indicating a substantial overlap in performance among advanced methods.
Computational Cost: The sorting step required for population updates incurs a non-negligible computational cost, potentially limiting applicability in time-constrained scenarios.
Hyperparameter Sensitivity: While the ECU reduces sensitivity, the algorithm still requires calibration of parameters like $r_{min}$ and $r_{max}$ , and performance can degrade if these are set outside recommended ranges.
Scope: The current implementation is a single-objective optimizer without formal theoretical convergence guarantees. The authors suggest future work should address high-dimensional problems ( $D > 500$ ), multi-objective optimization, and discrete search spaces.

The Big Idea: Your Algorithm is a Car Engine

The Brain: The "ECU" (Engine Control Unit)

Did It Work?

The "Ablation" Test (Taking it apart)

The Bottom Line

More like this