Approximating the solution of a dynamic, stochastic multiple knapsack problem