Innovative meta-heuristic method development based on uniform population and derivative

Optimization algorithms are used to search and find the best solution for a problem. Optimization algorithms, which are divided into two as exact and approximate methods (heuristic and metaheuristic), are observed to be used in all areas. Optimization algorithms basically consist of five parts (initial population generation, fitness computation, selection, creating a new generation and decision making). Researchers have made changes and improvements in these five sections with advancing technologies. The least work has been done in the initial population generation section. The random initiation method is still used the most today. Researchers have not done much research on this topic as it is considered a standard initiation method. In recent years, many new population initiation methods have been proposed to increase population diversity and uniform distribution. Within the scope of the thesis, initial population generation methods are examined in detail and a new categorization is proposed as a result of this review. In addition, method of creating a deterministic new initial population is suggested to literatüre. a linear function to represent the iris data set was obtained by making use of the multivariate linear regression (MLR) model initiated with this new initialization method. SGD, Momentum, Adagrad, RMSProp, Adadelta and Adam optimization algorithms were used to find the optimum values of coefficients of this function. In addtion, IAE, ITAE, MSE and ISE error functions were adopted as the objective function. First, initial populations of the methods were developed by using deterministic and stochastic initialization methods between upper and lower bounds. The method that was initialized stochasticaly was run several times as seen in literature and the mean values were calculated. On the other hand, the application that was initialized deterministic was only run once. According to deterministic and stochastic initialization Outputs, theta and iteration number were found to be close. However, temporal gain was achieved from the application that was initialized deterministic. Genarated outputs were compared and analyzed. According to comparisons, the linear model obtained using the Adadelta optimization algorithm and the MSE objective function performed best. Keywords: Deterministic Initial Population, Stochastic Initial Population, Multivariate Linear Regression, Optimization Algorithms