‎Development of a solution search method using an improved locust swarm algorithm

Abstract

The object of the research is decision support systems. The subject of the research is the decision-making process in management problems using the locust swarm algorithm and evolving artificial neural networks. A solution search method using an improved locust swarm algorithm is proposed. The research is based on the locust swarm algorithm for finding a solution regarding the state of an object. For training locust agents (LA), evolving artificial neural networks are used. The method has the following sequence of steps: - input of initial data; - processing of initial data taking into account the degree of uncertainty; - initial setting of LA in the search area; - determination of the initial speed of the LA movement; - a search vector is generated taking into account the degree of uncertainty; - calculation of the change in the value of the LA fitness function; - training of LA knowledge bases. The originality of the proposed method lies in the arrangement of LA taking into account the uncertainty of the initial data, improved procedures of global and local search taking into account the degree of noise of data about the state of the analysis object. Also, the originality of the research is avoiding the concentration of LA on the current best positions, reducing the probability of premature convergence of the algorithm and maintaining a balance between the convergence rate of the algorithm and diversification. The peculiarity of the proposed method is the use of an improved procedure for LA training. The training procedure consists in learning the parameters and architecture of individual elements and the architecture of the artificial neural network as a whole.