Abstract:This paper presents two simple yet powerful optimization algorithms named Best-Mean-Random (BMR) and Best-Worst-Randam (BWR) algorithms to handle both constrained and unconstrained optimization problems. These algorithms are free of metaphors and algorithm-specific parameters. The BMR algorithm is based on the best, mean, and random solutions of the population generated for solving a given problem; and the BWR algorithm is based on the best, worst, and random solutions. The performances of the proposed two algorithms are investigated by implementing them on 26 real-life non-convex constrained optimization problems given in the Congress on Evolutionary Computation (CEC) 2020 competition and comparisons are made with those of the other prominent optimization algorithms. Furthermore, computational experiments are conducted on 30 unconstrained standard benchmark optimization problems including 5 recently developed benchmark problems having distinct characteristics. The results proved the better competitiveness and superiority of the proposed simple algorithms. The optimization research community may gain an advantage by adapting these algorithms to solve various constrained and unconstrained real-life optimization problems across various scientific and engineering disciplines.
What problem does this paper attempt to address?
The main problems that this paper attempts to solve are to develop two simple optimization algorithms that are not metaphor - dependent in order to deal with non - convex constrained and unconstrained optimization problems in real - life. Specifically, the paper proposes two algorithms, Best - Mean - Random (BMR) and Best - Worst - Random (BWR), and verifies their effectiveness and superiority through a series of experiments.
### Core Problems of the Paper
1. **Avoid Using Metaphors**: Many existing optimization algorithms rely on metaphors from nature or other phenomena (such as animal behaviors, physical phenomena, etc.), but these metaphors are often irrelevant to the actual optimization process, resulting in a lack of scientific basis in algorithm design. The paper attempts to prove that effective optimization algorithms can be developed without relying on metaphors.
2. **Simplify Algorithm Parameters**: Traditional optimization algorithms usually need to adjust multiple parameters, which increases the complexity and uncertainty of use. The BMR and BWR algorithms aim to reduce or eliminate these algorithm - specific parameters, making the algorithms easier to use.
3. **Handle Complex Optimization Problems**: The algorithms in the paper are applicable not only to unconstrained optimization problems but also to non - convex optimization problems with complex constraints. The author verifies the effectiveness of the algorithms by testing on 26 non - convex constrained optimization problems in the CEC 2020 competition and 30 standard unconstrained benchmark optimization problems.
### Specific Objectives
- **Prove Independence from Metaphors**: Demonstrate that effective optimization algorithms can be constructed through simple mathematical models without introducing complex metaphors.
- **Develop Simple Optimization Algorithms without Specific Parameters**: Design two algorithms, BMR and BWR, to make them easy to understand and apply.
- **Verify the Performance of the Algorithms in Practical Problems**: Through a large number of experiments, prove the high efficiency and competitiveness of the BMR and BWR algorithms in dealing with non - convex constrained and unconstrained optimization problems.
### Experimental Verification
The paper shows the performance of the BMR and BWR algorithms through experiments on 26 non - convex constrained optimization problems and 30 standard unconstrained optimization problems in the CEC 2020 competition. The experimental results indicate that these two algorithms perform excellently in terms of convergence speed and result quality, and are even superior to other well - known optimization algorithms.
### Conclusion
This paper successfully proposes a new design concept for optimization algorithms, that is, not relying on metaphors and simplifying parameter settings, and at the same time verifies its effectiveness in practical problems. This method provides new ideas for research in the field of optimization and helps promote the development of optimization algorithms.
### Related Formulas
In the BMR algorithm, the variable update formula is as follows:
\[
V'_{j,k,i} =
\begin{cases}
V_{j,k,i} + r_{1,j,i} (V_{j,\text{best},i} - T \cdot V_{j,\text{mean},i}) + r_{2,j,i} (V_{j,\text{best},i} - V_{j,\text{random},i}), & \text{if } r_4 > 0.5 \\
U_j - (U_j - L_j)r_3, & \text{otherwise}
\end{cases}
\]
In the BWR algorithm, the variable update formula is as follows:
\[
V'_{j,k,i} =
\begin{cases}
V_{j,k,i} + r_{1,j,i} (V_{j,\text{best},i} - T \cdot V_{j,\text{random},i}) - r_{2,j,i} (V_{j,\text{worst},i} - V_{j,\text{random},i}), & \text{if } r_4 > 0.5 \\
U_j - (U_j - L_j)r_3, & \text{otherwise}
\end{cases}
\]
where:
- \( r_1, r_2, r_3, r_4 \sim \text{Uniform}(0, 1) \)
- \( T \sim \text{Choice}(\{1, 2\}) \)