Parallel implementation of simulated annealing by distributed memory systems

The aim of this paper is to provide a parallel strategy to be used when implementing the Simulated Annealing process by distributed memory systems. It focuses the attention on the portfolio selection problem with dynamic and integer constraints on the variables. However, this method can be extended to any dynamic optimisation problem as the optimal control. In the first part it develops the sequential algorithm of Simulated Annealing stressing some features inherent in the application itself: the structure of the neighbourhood of the solution, the starting value and the decreasing schedule of the control parameter. In the second part it deals with the formulation of a parallel strategy for the algorithm. It starts from the consideration that an optimisation technique with a probabilistic searching criterion cannot give a reliable result in a single execution. Therefore it proposes a strategy performing a number of independent concurrent Simulated Annealing processes applied to different initial configurations of the same problem. Finally, it proposes a co-operation policy among processes, by means of a reduced number of communications. This policy enhance the efficiency of the parallel computing system. INTRODUCTION Investment diversification is known as the best way to reduce the intrinsic risk of individual financial assets. Transactions on Information and Communications Technologies vol 3, © 1993 WIT Press, www.witpress.com, ISSN 1743-3517