Doğadan esinlenen evrimsel algoritmalar ve bir model önerisi

This study introduces a new population-based evolutionary computing model for solving linear/nonlinear continuous unconstrained/constrained optimization problems. The proposed model includes two optimization algorithms. The first one is an initialization algorithm that provides adaptive initial solutions, to some extent, reducing the diversity of randomness in the initialization of the algorithms for problems that may have many local optimums. The prominent feature of the algorithm is the ability to narrow the search space adaptively without falling into local optimums and changing the nature of the problem. Unlike simple random approaches, the proposed algorithm escapes from inadvertently removing the global optimum in multi-modal problems. In terms of time and performance, the initialization algorithm doesn't add additional burden, on the contrary, it contributes to the problem-solving procedure. The second proposed algorithm called Repulsive Forces Optimization (REF) depends on Newton's General Gravity Law and Coulomb's Law. Different from the first algorithm, the REF algorithm aims to reach optimum-like solutions by constraint-handling abilities. REF algorithm assumes that likely charged particles in a bounded space are possible solution points. The forces between the particle and its neighbors make the particle moved to a new location where a better solution may exist. The repulsive structure of the particles could be considered as the mimics of Coulomb's Law. Furthermore, Tabu Search Algorithm and Elitism selection approach inspire the memory usage of the proposed algorithm. The inspirations of the REF algorithm are determined to create the best combination of features that provides better results. Besides, this algorithm is structured on the principle of multiplicative penalty approach that considers satisfaction rates and the total deviations of constraints as well as objective function value for constraint handling. For this reason, it can handle continuous constrained problems very well. The performances of the algorithms are evaluated with unconstrained/bounded optimization benchmarks and engineering design problems that belong to the most commonly used cases by evolutionary optimization researchers. In addition, an economic dispatch problem is also applied for benchmarking. It is concluded that the initialization algorithm converges much better than random solutions and it is applicable for further studies focusing on better initial solutions that guide reaching an optimal solution. Experimental results of real-world problems show that the proposed algorithms produce satisfactory results compared to the methods published in the literature.