Abstract
To tackle the shortcoming of deficient using of feedback information in quantum-inspired evolutionary computing and catastrophe such as quantum mutation, the improved quantum-inspired evolutionary algorithm (IQEA) is proposed. The crossover and mutation operators of genetic algorithm are introduced to the quantum evolution algorithm, at the same time, the better individuals information is applied to the generation of the population applying the probability evaluation model of the estimation of distribution algorithm. Theory analyze and test function simulation experiments show that the improved quantum genetic algorithm is characterized by rapid convergence, excellent robustness and so on.
Introduction
Evolutionary algorithms (EAs) are essential tools for solving complex optimization problems where traditional gradient-based methods are ineffective. However, standard Quantum-inspired Evolutionary Algorithms (QEAs) often suffer from two main issues: deficient feedback—failing to effectively use information from previous generations—and catastrophic stagnation, where they get stuck in local optima due to a lack of population diversity.
The Improved Quantum-inspired Evolutionary Algorithm (IQEA) addresses these issues by bridging the gap between "micro-level" individual manipulation and "macro-level" population modeling. By combining the probabilistic power of quantum computing with the structural strengths of classical genetic algorithms and Estimation of Distribution Algorithms (EDAs), IQEA achieves significantly faster convergence and superior global search capabilities.
Methodology
The IQEA architecture is built upon three primary pillars:
- Quantum Representation: Individuals are represented as "qubits" using probability amplitudes ($\alpha$ and $\beta$). This allows a single individual to represent a superposition of many states simultaneously, vastly increasing search space coverage.
- Classical Genetic Operators: Crossover and mutation operators from classical Genetic Algorithms (GA) are integrated into the quantum framework to maintain population diversity and explore new regions of the solution space.
- Estimation of Distribution (EDA) Integration: The algorithm employs a probability evaluation model to analyze high-performing individuals. This "macro" information guides the generation of the next population, ensuring the search follows promising global trends rather than random fluctuations.
In the IQEA loop, half the population is updated using standard quantum gates, while the other half is updated based on the EDA’s probability model of the best-performing individuals.
Experimental Setup
The researchers evaluated IQEA using the De Jong function (specifically the Rosenbrock variant), a standard mathematical benchmark for optimization. This non-linear, two-dimensional function is notoriously difficult to maximize because its global peak resides in a narrow, curved valley. IQEA's performance was compared against two baselines: the standard Genetic Algorithm (GA) and the basic Quantum-inspired Genetic Algorithm (QGA).
Results and Discussion
The simulation results, based on 50 independent runs to ensure statistical reliability, demonstrate the clear advantages of the IQEA approach.
As shown in the convergence plot, IQEA (solid blue line) reaches the global maximum significantly faster than both the standard GA and the QGA. While the classical GA evolved the slowest and frequently failed to reach the true global optimum, IQEA consistently found the peak ($f \approx 3905.926$).
The integration of micro-level manipulation (crossover/mutation) and macro-level modeling (EDA) proved complementary. The qubit representation allowed the algorithm to effectively "hedge its bets" across multiple potential solutions, while the EDA component learned from a pool of "better" individuals to preserve the genetic diversity needed to avoid local optima. Furthermore, IQEA exhibited excellent robustness, performing consistently across different runs despite the inherent randomness of evolutionary processes.
Conclusion and Future Work
The IQEA demonstrates a powerful hybrid approach to optimization, though some areas remain for future exploration. The computational overhead of building the EDA probability model in every generation should be assessed for high-dimensional problems. Additionally, future benchmarking on a wider variety of NP-hard combinatorial problems and real-world engineering constraints would further validate the algorithm's versatility. Finally, optimizing the ratio between quantum gate updates and EDA updates could potentially yield even greater performance gains.
Summary updated