Function optimisation by learning automata

This paper presents a new algorithm, Function Optimisation by Learning Automata (FOLA), to solve complex function optimisation problems. FOLA consists of multiple automata, in which each automaton undertakes dimensional search on a selected dimension of the solution domain. A search action is taken on a path which is identified in the search space by the path value, and the path value is updated using the values of the states visited in the past, via a state memory that enables better use of the information collected in the optimisation process. In this paper, FOLA is compared with two popularly used particle swarm optimisers and four newly-proposed optimisers, on nine complex multi-modal benchmark functions. The experimental results have shown that in comparison with the other optimisers, FOLA offers better performance for most of the benchmark functions, in terms of its convergence rate and accuracy, and it uses much less computation time to obtain accurate solutions, especially for high-dimensional functions. In order to explore the FOLA’s potential for applications, it is also applied to solve an optimal power flow problem of power systems. FOLA is able to minimise the fuel cost and enhance the voltage stability of the power system more efficiently in comparison with the other algorithms.