Data Mining Approach to Select Parameters of Swarm Intelligence Algorithms for Optimal Placement Reactive Power Compensation

Abstract

This study considers the problem of the parameter’s selection of the Particle swarm optimization algorithm. To find effective values of the algorithm parameters, the regression analysis, and rule-based classification have been applied. The regression analysis has allowed creating the model of dependency between values of the algorithm parameters and efficiency of solutions achieved. The classifier has approved allocating areas of the most effective values of the parameters. The areas obtained were used as the areas of possible values when searching extremum of regression curves per each of parameters. The effectiveness of parameters obtained has been compared with the parameters recommended by other researchers. It was found out that using the regression analysis and rule-based classifier allowed to select values of Particle swarm optimization algorithm parameters successfully. An example of using the method to solve the problem of optimal placement of reactive power compensation units in the electrical network is given.